Classification of Twitter users with eating disorder engagement : learning from the biographies
- Abuhassan, Mohammad, Anwar, Tarique, Fuller-Tyszkiewicz, Matthew, Jarman, Hannah, Shatte, Adrian, Liu, Chengfei, Sukunesan, Suku
- Authors: Abuhassan, Mohammad , Anwar, Tarique , Fuller-Tyszkiewicz, Matthew , Jarman, Hannah , Shatte, Adrian , Liu, Chengfei , Sukunesan, Suku
- Date: 2023
- Type: Text , Journal article
- Relation: Computers in Human Behavior Vol. 140, no. (2023), p.
- Full Text: false
- Reviewed:
- Description: Individuals with an Eating Disorder (ED) are typically reluctant to seek help via traditional means (e.g., psychologists). However, recent evidence suggests that many individuals seek assistance via social media for weight and diet related concerns. Sophisticated approaches are needed to better distinguish those who may be in need of help for an ED from those who are simply commenting on ED in online social environments. In order to facilitate effective communication between individuals with or at-risk of an ED and healthcare professionals, this research exploits a deep learning model to differentiate the users with ED engagement (e.g., ED sufferers, healthcare professionals or communicators) over social media. For this purpose, a collection of Twitter data is compiled using Twitter application programming interface (API) on the Australian Research Data Commons (ARDC) Nectar research cloud. After collecting 1,400,000 Twitter biographies in total, a subset of 4000 biographies are annotated manually. This annotation enables the differentiation of users engaged with ED-focused language on social media into five categories: ED-user, healthcare professional, communicator, healthcare professional-communicator, and other. Based on these annotated categories, a predictive deep learning model based on bidirectional encoder representations from transformers (BERT) and long short-term memory (LSTM) is developed. The model achieves an F1 score of 98.19% and an accuracy of 98.37%. It demonstrates the viability of detecting the individuals with possible ED risk and distinguishes them from other categories using their biography data. We further conducted a network analysis for investigating the communication network between these categories. Our analysis shows that ED-users are more secretive and self-protective, whereas the healthcare professionals and communicators frequently interact with each other and a wide range of other people. To the best of our knowledge, our research is the first of its kind for identifying the different user categories engaged with ED-focused communications on social media. © 2022
- Ali, Farman, El-Sappagh, Shaker, Islam, S., Kwak, Daehan, Ali, Amjad, Imran, Muhammad, Kwak, Kyung-Sup
- Authors: Ali, Farman , El-Sappagh, Shaker , Islam, S. , Kwak, Daehan , Ali, Amjad , Imran, Muhammad , Kwak, Kyung-Sup
- Date: 2020
- Type: Text , Journal article
- Relation: Information Fusion Vol. 63, no. (2020), p. 208-222
- Full Text: false
- Reviewed:
- Description: The accurate prediction of heart disease is essential to efficiently treating cardiac patients before a heart attack occurs. This goal can be achieved using an optimal machine learning model with rich healthcare data on heart diseases. Various systems based on machine learning have been presented recently to predict and diagnose heart disease. However, these systems cannot handle high-dimensional datasets due to the lack of a smart framework that can use different sources of data for heart disease prediction. In addition, the existing systems utilize conventional techniques to select features from a dataset and compute a general weight for them based on their significance. These methods have also failed to enhance the performance of heart disease diagnosis. In this paper, a smart healthcare system is proposed for heart disease prediction using ensemble deep learning and feature fusion approaches. First, the feature fusion method combines the extracted features from both sensor data and electronic medical records to generate valuable healthcare data. Second, the information gain technique eliminates irrelevant and redundant features, and selects the important ones, which decreases the computational burden and enhances the system performance. In addition, the conditional probability approach computes a specific feature weight for each class, which further improves system performance. Finally, the ensemble deep learning model is trained for heart disease prediction. The proposed system is evaluated with heart disease data and compared with traditional classifiers based on feature fusion, feature selection, and weighting techniques. The proposed system obtains accuracy of 98.5%, which is higher than existing systems. This result shows that our system is more effective for the prediction of heart disease, in comparison to other state-of-the-art methods. © 2020
- Ali, Sajid, El-Sappagh, Shaker, Ali, Farman, Imran, Muhammad, Abuhmed, Tamer
- Authors: Ali, Sajid , El-Sappagh, Shaker , Ali, Farman , Imran, Muhammad , Abuhmed, Tamer
- Date: 2022
- Type: Text , Journal article
- Relation: IEEE Journal of Biomedical and Health Informatics Vol. 26, no. 12 (2022), p. 5793-5804
- Full Text: false
- Reviewed:
- Description: In a hospital, accurate and rapid mortality prediction of Length of Stay (LOS) is essential since it is one of the essential measures in treating patients with severe diseases. When predictions of patient mortality and readmission are combined, these models gain a new level of significance. Therefore, the most expensive components of patient care are LOS and readmission rates. Several studies have assessed readmission to the hospital as a single-task issue. The performance, robustness, and stability of the model increase when many correlated tasks are optimized. This study develops multimodal multitasking Long Short-Term Memory (LSTM) Deep Learning (DL) model that can predict both LOS and readmission for patients using multi-sensory data from 47 patients. Continuous sensory data is divided into eight sections, each of which is recorded for an hour. The time steps are constructed using a dual 10-second window-based technique, resulting in six steps per hour. The 30 statistical features are computed by transforming the sensory input into the resulting vector. The proposed multitasking model predicts 30-day readmission as a binary classification problem and LOS as a regression task by constructing discrete time-step data based on the length of physical activity during a hospital stay. The proposed model is compared to a random forest for a single-task problem (classification or regression) because typical machine learning algorithms are unable to handle the multitasking challenge. In addition, sensory data combined with other cost-effective modalities such as demographics, laboratory tests, and comorbidities to construct reliable models for personalized, cost-effective, and medically acceptable prediction. With a high accuracy of 94.84%, the proposed multitask multimodal DL model classifies the patient's readmission status and determines the patient's LOS in hospital with a minimal Mean Square Error (MSE) of 0.025 and Root Mean Square Error (RMSE) of 0.077, which is promising, effective, and trustworthy. © 2013 IEEE.
Deep learning and big data technologies for IoT security
- Amanullah, Mohamed, Habeeb, Riyaz, Nasaruddin, Fariza, Gani, Abdullah, Ahmed, Ejaz, Nainar, Abdul, Akim, Nazihah, Imran, Muhammad
- Authors: Amanullah, Mohamed , Habeeb, Riyaz , Nasaruddin, Fariza , Gani, Abdullah , Ahmed, Ejaz , Nainar, Abdul , Akim, Nazihah , Imran, Muhammad
- Date: 2020
- Type: Text , Journal article , Review
- Relation: Computer Communications Vol. 151, no. (2020), p. 495-517
- Full Text: false
- Reviewed:
- Description: Technology has become inevitable in human life, especially the growth of Internet of Things (IoT), which enables communication and interaction with various devices. However, IoT has been proven to be vulnerable to security breaches. Therefore, it is necessary to develop fool proof solutions by creating new technologies or combining existing technologies to address the security issues. Deep learning, a branch of machine learning has shown promising results in previous studies for detection of security breaches. Additionally, IoT devices generate large volumes, variety, and veracity of data. Thus, when big data technologies are incorporated, higher performance and better data handling can be achieved. Hence, we have conducted a comprehensive survey on state-of-the-art deep learning, IoT security, and big data technologies. Further, a comparative analysis and the relationship among deep learning, IoT security, and big data technologies have also been discussed. Further, we have derived a thematic taxonomy from the comparative analysis of technical studies of the three aforementioned domains. Finally, we have identified and discussed the challenges in incorporating deep learning for IoT security using big data technologies and have provided directions to future researchers on the IoT security aspects. © 2020 Elsevier B.V.
Efficient future waste management : a learning-based approach with deep neural networks for smart system (LADS)
- Chauhan, Ritu, Shighra, Sahil, Madkhali, Hatim, Nguyen, Linh, Prasad, Mukesh
- Authors: Chauhan, Ritu , Shighra, Sahil , Madkhali, Hatim , Nguyen, Linh , Prasad, Mukesh
- Date: 2023
- Type: Text , Journal article
- Relation: Applied Sciences (Switzerland) Vol. 13, no. 7 (2023), p.
- Full Text:
- Reviewed:
- Description: Waste segregation, management, transportation, and disposal must be carefully managed to reduce the danger to patients, the public, and risks to the environment’s health and safety. The previous method of monitoring trash in strategically placed garbage bins is a time-consuming and inefficient method that wastes time, human effort, and money, and is also incompatible with smart city needs. So, the goal is to reduce individual decision-making and increase the productivity of the waste categorization process. Using a convolutional neural network (CNN), the study sought to create an image classifier that recognizes items and classifies trash material. This paper provides an overview of trash monitoring methods, garbage disposal strategies, and the technology used in establishing a waste management system. Finally, an efficient system and waste disposal approach is provided that may be employed in the future to improve performance and cost effectiveness. One of the most significant barriers to efficient waste management can now be overcome with the aid of a deep learning technique. The proposed method outperformed the alternative AlexNet, VGG16, and ResNet34 methods. © 2023 by the authors.
- Authors: Chauhan, Ritu , Shighra, Sahil , Madkhali, Hatim , Nguyen, Linh , Prasad, Mukesh
- Date: 2023
- Type: Text , Journal article
- Relation: Applied Sciences (Switzerland) Vol. 13, no. 7 (2023), p.
- Full Text:
- Reviewed:
- Description: Waste segregation, management, transportation, and disposal must be carefully managed to reduce the danger to patients, the public, and risks to the environment’s health and safety. The previous method of monitoring trash in strategically placed garbage bins is a time-consuming and inefficient method that wastes time, human effort, and money, and is also incompatible with smart city needs. So, the goal is to reduce individual decision-making and increase the productivity of the waste categorization process. Using a convolutional neural network (CNN), the study sought to create an image classifier that recognizes items and classifies trash material. This paper provides an overview of trash monitoring methods, garbage disposal strategies, and the technology used in establishing a waste management system. Finally, an efficient system and waste disposal approach is provided that may be employed in the future to improve performance and cost effectiveness. One of the most significant barriers to efficient waste management can now be overcome with the aid of a deep learning technique. The proposed method outperformed the alternative AlexNet, VGG16, and ResNet34 methods. © 2023 by the authors.
LSPM : joint deep modeling of long-term preference and short-term preference for recommendation
- Chen, Jie, Jiang, Lifen, Sun, Huazhi, Ma, Chunmei, Liu, Zekang, Zhao, Dake
- Authors: Chen, Jie , Jiang, Lifen , Sun, Huazhi , Ma, Chunmei , Liu, Zekang , Zhao, Dake
- Date: 2019
- Type: Text , Conference paper
- Relation: 26th International Conference on Neural Information Processing, ICONIP 2019 Vol. 1142 CCIS, p. 237-246
- Full Text: false
- Reviewed:
- Description: In the era of information, recommender systems are playing an indispensable role in our lives. A lot of deep learning based recommender systems have been created and proven to be good progress. However, users’ decisions are determined by both long-term and short-term preferences, and most of the existing efforts study these two requirements separately. In this paper, we seek to build a bridge between the long-term and short-term preferences. We propose a Long & Short-term Preference Model (LSPM), which incorporates LSTM and self-attention mechanism to learn the short-term preference and jointly model the long-term preference by a neural latent factor model. We conduct experiments to demonstrate the effectiveness of LSPM on three public datasets. Compared with the state-of-the-art methods, LSPM got a significant improvement in HR@10 and NDCG@10, which relatively increased by 3.875% and 6.363%. We publish our code at https://github.com/chenjie04/LSPM/. © Springer Nature Switzerland AG 2019.
Subgraph adaptive structure-aware graph contrastive learning
- Chen, Zhikui, Peng, Yin, Yu, Shuo, Cao, Chen, Xia, Feng
- Authors: Chen, Zhikui , Peng, Yin , Yu, Shuo , Cao, Chen , Xia, Feng
- Date: 2022
- Type: Text , Journal article
- Relation: Mathematics (Basel) Vol. 10, no. 17 (2022), p. 3047
- Full Text:
- Reviewed:
- Description: Graph contrastive learning (GCL) has been subject to more attention and been widely applied to numerous graph learning tasks such as node classification and link prediction. Although it has achieved great success and even performed better than supervised methods in some tasks, most of them depend on node-level comparison, while ignoring the rich semantic information contained in graph topology, especially for social networks. However, a higher-level comparison requires subgraph construction and encoding, which remain unsolved. To address this problem, we propose a subgraph adaptive structure-aware graph contrastive learning method (PASCAL) in this work, which is a subgraph-level GCL method. In PASCAL, we construct subgraphs by merging all motifs that contain the target node. Then we encode them on the basis of motif number distribution to capture the rich information hidden in subgraphs. By incorporating motif information, PASCAL can capture richer semantic information hidden in local structures compared with other GCL methods. Extensive experiments on six benchmark datasets show that PASCAL outperforms state-of-art graph contrastive learning and supervised methods in most cases.
- Authors: Chen, Zhikui , Peng, Yin , Yu, Shuo , Cao, Chen , Xia, Feng
- Date: 2022
- Type: Text , Journal article
- Relation: Mathematics (Basel) Vol. 10, no. 17 (2022), p. 3047
- Full Text:
- Reviewed:
- Description: Graph contrastive learning (GCL) has been subject to more attention and been widely applied to numerous graph learning tasks such as node classification and link prediction. Although it has achieved great success and even performed better than supervised methods in some tasks, most of them depend on node-level comparison, while ignoring the rich semantic information contained in graph topology, especially for social networks. However, a higher-level comparison requires subgraph construction and encoding, which remain unsolved. To address this problem, we propose a subgraph adaptive structure-aware graph contrastive learning method (PASCAL) in this work, which is a subgraph-level GCL method. In PASCAL, we construct subgraphs by merging all motifs that contain the target node. Then we encode them on the basis of motif number distribution to capture the rich information hidden in subgraphs. By incorporating motif information, PASCAL can capture richer semantic information hidden in local structures compared with other GCL methods. Extensive experiments on six benchmark datasets show that PASCAL outperforms state-of-art graph contrastive learning and supervised methods in most cases.
- Chiang, Christina, Wells, Paul, Fieger, Peter, Sharma, Divesh
- Authors: Chiang, Christina , Wells, Paul , Fieger, Peter , Sharma, Divesh
- Date: 2021
- Type: Text , Journal article
- Relation: Accounting and Finance Vol. 61, no. 1 (2021), p. 913-936
- Full Text: false
- Reviewed:
- Description: Arguably, the audit course is one of the most challenging as it links prior accounting knowledge with new audit knowledge that students are generally not exposed to. A mini-audit group project was implemented at a New Zealand university, and a learning approach and learning experience survey instrument was administered. Responses from 98 students suggest that they perceived the learning experience positively and were encouraged to adopt a deep approach to learning. The findings have implications for accounting educators in the design and development of learning and assessment strategies in an audit course. © 2020 Accounting and Finance Association of Australia and New Zealand
Comparative analysis of machine and deep learning models for soil properties prediction from hyperspectral visual band
- Datta, Dristi, Paul, Manoranjan, Murshed, Manzur, Teng, Shyh Wei, Schmidtke, Leigh
- Authors: Datta, Dristi , Paul, Manoranjan , Murshed, Manzur , Teng, Shyh Wei , Schmidtke, Leigh
- Date: 2023
- Type: Text , Journal article
- Relation: Environments Vol. 10, no. 5 (2023), p. 77
- Full Text:
- Reviewed:
- Description: Estimating various properties of soil, including moisture, carbon, and nitrogen, is crucial for studying their correlation with plant health and food production. However, conventional methods such as oven-drying and chemical analysis are laborious, expensive, and only feasible for a limited land area. With the advent of remote sensing technologies like multi/hyperspectral imaging, it is now possible to predict soil properties non-invasive and cost-effectively for a large expanse of bare land. Recent research shows the possibility of predicting those soil contents from a wide range of hyperspectral data using good prediction algorithms. However, these kinds of hyperspectral sensors are expensive and not widely available. Therefore, this paper investigates different machine and deep learning techniques to predict soil nutrient properties using only the red (R), green (G), and blue (B) bands data to propose a suitable machine/deep learning model that can be used as a rapid soil test. Another objective of this research is to observe and compare the prediction accuracy in three cases i. hyperspectral band ii. full spectrum of the visual band, and iii. three-channel of RGB band and provide a guideline to the user on which spectrum information they should use to predict those soil properties. The outcome of this research helps to develop a mobile application that is easy to use for a quick soil test. This research also explores learning-based algorithms with significant feature combinations and their performance comparisons in predicting soil properties from visual band data. For this, we also explore the impact of dimensional reduction (i.e., principal component analysis) and transformations (i.e., empirical mode decomposition) of features. The results show that the proposed model can comparably predict the soil contents from the three-channel RGB data.
- Authors: Datta, Dristi , Paul, Manoranjan , Murshed, Manzur , Teng, Shyh Wei , Schmidtke, Leigh
- Date: 2023
- Type: Text , Journal article
- Relation: Environments Vol. 10, no. 5 (2023), p. 77
- Full Text:
- Reviewed:
- Description: Estimating various properties of soil, including moisture, carbon, and nitrogen, is crucial for studying their correlation with plant health and food production. However, conventional methods such as oven-drying and chemical analysis are laborious, expensive, and only feasible for a limited land area. With the advent of remote sensing technologies like multi/hyperspectral imaging, it is now possible to predict soil properties non-invasive and cost-effectively for a large expanse of bare land. Recent research shows the possibility of predicting those soil contents from a wide range of hyperspectral data using good prediction algorithms. However, these kinds of hyperspectral sensors are expensive and not widely available. Therefore, this paper investigates different machine and deep learning techniques to predict soil nutrient properties using only the red (R), green (G), and blue (B) bands data to propose a suitable machine/deep learning model that can be used as a rapid soil test. Another objective of this research is to observe and compare the prediction accuracy in three cases i. hyperspectral band ii. full spectrum of the visual band, and iii. three-channel of RGB band and provide a guideline to the user on which spectrum information they should use to predict those soil properties. The outcome of this research helps to develop a mobile application that is easy to use for a quick soil test. This research also explores learning-based algorithms with significant feature combinations and their performance comparisons in predicting soil properties from visual band data. For this, we also explore the impact of dimensional reduction (i.e., principal component analysis) and transformations (i.e., empirical mode decomposition) of features. The results show that the proposed model can comparably predict the soil contents from the three-channel RGB data.
- George, Yasmeen, Antony, Bhavna, Ishikawa, Hiroshi, Wollstein, Gadi, Schuman, Joel, Garnavi, Rahil
- Authors: George, Yasmeen , Antony, Bhavna , Ishikawa, Hiroshi , Wollstein, Gadi , Schuman, Joel , Garnavi, Rahil
- Date: 2020
- Type: Text , Journal article
- Relation: IEEE journal of biomedical and health informatics Vol. 24, no. 12 (2020), p. 3421-3430
- Full Text: false
- Reviewed:
- Description: The direct analysis of 3D Optical Coherence Tomography (OCT) volumes enables deep learning models (DL) to learn spatial structural information and discover new bio-markers that are relevant to glaucoma. Downsampling 3D input volumes is the state-of-art solution to accommodate for the limited number of training volumes as well as the available computing resources. However, this limits the network's ability to learn from small retinal structures in OCT volumes. In this paper, our goal is to improve the performance by providing guidance to DL model during training in order to learn from finer ocular structures in 3D OCT volumes. Therefore, we propose an end-to-end attention guided 3D DL model for glaucoma detection and estimating visual function from retinal structures. The model consists of three pathways with the same network architecture but different inputs. One input is the original 3D-OCT cube and the other two are computed during training guided by the 3D gradient class activation heatmaps. Each pathway outputs the class-label and the whole model is trained concurrently to minimize the sum of losses from three pathways. The final output is obtained by fusing the predictions of the three pathways. Also, to explore the robustness and generalizability of the proposed model, we apply the model on a classification task for glaucoma detection as well as a regression task to estimate visual field index (VFI) (a value between 0 and 100). A 5-fold cross-validation with a total of 3782 and 10,370 OCT scans is used to train and evaluate the classification and regression models, respectively. The glaucoma detection model achieved an area under the curve (AUC) of 93.8% compared with 86.8% for a baseline model without the attention-guided component. The model also outperformed six different feature based machine learning approaches that use scanner computed measurements for training. Further, we also assessed the contribution of different retinal layers that are relevant to glaucoma. The VFI estimation model achieved a Pearson correlation and median absolute error of 0.75 and 3.6%, respectively, for a test set of size 3100 cubes.
Retinal optical coherence tomography image enhancement via deep learning
- Halupka, Kerry, Antony, Bhavna, Lee, Matthew, Lucy, Katie, Rai, Ravneet, Ishikawa, Hiroshi, Wollstein, Gadi, Schuman, Joel, Garnavi, Rahil
- Authors: Halupka, Kerry , Antony, Bhavna , Lee, Matthew , Lucy, Katie , Rai, Ravneet , Ishikawa, Hiroshi , Wollstein, Gadi , Schuman, Joel , Garnavi, Rahil
- Date: 2018
- Type: Text , Journal article
- Relation: Biomedical Optics Express Vol. 9, no. 12 (2018), p. 6205-6221
- Full Text: false
- Reviewed:
- Description: Optical coherence tomography (OCT) images of the retina are a powerful tool for diagnosing and monitoring eye disease. However, they are plagued by speckle noise, which reduces image quality and reliability of assessment. This paper introduces a novel speckle reduction method inspired by the recent successes of deep learning in medical imaging. We present two versions of the network to reflect the needs and preferences of different end-users. Specifically, we train a convolution neural network to denoise cross-sections from OCT volumes of healthy eyes using either (1) mean-squared error, or (2) a generative adversarial network (GAN) with Wasserstein distance and perceptual similarity. We then interrogate the success of both methods with extensive quantitative and qualitative metrics on cross-sections from both healthy and glaucomatous eyes. The results show that the former approach provides state-of-the-art improvement in quantitative metrics such as PSNR and SSIM, and aids layer segmentation. However, the latter approach, which puts more weight on visual perception, outperformed for qualitative comparisons based on accuracy, clarity, and personal preference. Overall, our results demonstrate the effectiveness and efficiency of a deep learning approach to denoising OCT images, while maintaining subtle details in the images.
A framework for cardiac arrhythmia detection from IoT-based ECGs
- He, Jinyuan, Rong, Jia, Sun, Le, Wang, Hua, Zhang, Yanchun, Ma, Jiangang
- Authors: He, Jinyuan , Rong, Jia , Sun, Le , Wang, Hua , Zhang, Yanchun , Ma, Jiangang
- Date: 2020
- Type: Text , Journal article
- Relation: World Wide Web Vol. 23, no. 5 (2020), p. 2835-2850
- Full Text:
- Reviewed:
- Description: Cardiac arrhythmia has been identified as a type of cardiovascular diseases (CVDs) that causes approximately 12% of all deaths globally. The development of Internet-of-Things has spawned novel ways for heart monitoring but also presented new challenges for manual arrhythmia detection. An automated method is highly demanded to provide support for physicians. Current attempts for automatic arrhythmia detection can roughly be divided as feature-engineering based and deep-learning based methods. Most of the feature-engineering based methods are suffering from adopting single classifier and use fixed features for classifying all five types of heartbeats. This introduces difficulties in identification of the problematic heartbeats and limits the overall classification performance. The deep-learning based methods are usually not evaluated in a realistic manner and report overoptimistic results which may hide potential limitations of the models. Moreover, the lack of consideration of frequency patterns and the heart rhythms can also limit the model performance. To fill in the gaps, we propose a framework for arrhythmia detection from IoT-based ECGs. The framework consists of two modules: a data cleaning module and a heartbeat classification module. Specifically, we propose two solutions for the heartbeat classification task, namely Dynamic Heartbeat Classification with Adjusted Features (DHCAF) and Multi-channel Heartbeat Convolution Neural Network (MCHCNN). DHCAF is a feature-engineering based approach, in which we introduce dynamic ensemble selection (DES) technique and develop a result regulator to improve classification performance. MCHCNN is deep-learning based solution that performs multi-channel convolutions to capture both temporal and frequency patterns from heartbeat to assist the classification. We evaluate the proposed framework with DHCAF and with MCHCNN on the well-known MIT-BIH-AR database, respectively. The results reported in this paper have proven the effectiveness of our framework. © 2020, Springer Science+Business Media, LLC, part of Springer Nature.
- Authors: He, Jinyuan , Rong, Jia , Sun, Le , Wang, Hua , Zhang, Yanchun , Ma, Jiangang
- Date: 2020
- Type: Text , Journal article
- Relation: World Wide Web Vol. 23, no. 5 (2020), p. 2835-2850
- Full Text:
- Reviewed:
- Description: Cardiac arrhythmia has been identified as a type of cardiovascular diseases (CVDs) that causes approximately 12% of all deaths globally. The development of Internet-of-Things has spawned novel ways for heart monitoring but also presented new challenges for manual arrhythmia detection. An automated method is highly demanded to provide support for physicians. Current attempts for automatic arrhythmia detection can roughly be divided as feature-engineering based and deep-learning based methods. Most of the feature-engineering based methods are suffering from adopting single classifier and use fixed features for classifying all five types of heartbeats. This introduces difficulties in identification of the problematic heartbeats and limits the overall classification performance. The deep-learning based methods are usually not evaluated in a realistic manner and report overoptimistic results which may hide potential limitations of the models. Moreover, the lack of consideration of frequency patterns and the heart rhythms can also limit the model performance. To fill in the gaps, we propose a framework for arrhythmia detection from IoT-based ECGs. The framework consists of two modules: a data cleaning module and a heartbeat classification module. Specifically, we propose two solutions for the heartbeat classification task, namely Dynamic Heartbeat Classification with Adjusted Features (DHCAF) and Multi-channel Heartbeat Convolution Neural Network (MCHCNN). DHCAF is a feature-engineering based approach, in which we introduce dynamic ensemble selection (DES) technique and develop a result regulator to improve classification performance. MCHCNN is deep-learning based solution that performs multi-channel convolutions to capture both temporal and frequency patterns from heartbeat to assist the classification. We evaluate the proposed framework with DHCAF and with MCHCNN on the well-known MIT-BIH-AR database, respectively. The results reported in this paper have proven the effectiveness of our framework. © 2020, Springer Science+Business Media, LLC, part of Springer Nature.
Automatic driver distraction detection using deep convolutional neural networks
- Hossain, Md Uzzol, Rahman, Md Ataur, Islam, Md Manowarul, Akhter, Arnisha, Uddin, Md Ashraf, Paul, Bikash
- Authors: Hossain, Md Uzzol , Rahman, Md Ataur , Islam, Md Manowarul , Akhter, Arnisha , Uddin, Md Ashraf , Paul, Bikash
- Date: 2022
- Type: Text , Journal article
- Relation: Intelligent Systems with Applications Vol. 14, no. (2022), p.
- Full Text:
- Reviewed:
- Description: Recently, the number of road accidents has been increased worldwide due to the distraction of the drivers. This rapid road crush often leads to injuries, loss of properties, even deaths of the people. Therefore, it is essential to monitor and analyze the driver's behavior during the driving time to detect the distraction and mitigate the number of road accident. To detect various kinds of behavior like- using cell phone, talking to others, eating, sleeping or lack of concentration during driving; machine learning/deep learning can play significant role. However, this process may need high computational capacity to train the model by huge number of training dataset. In this paper, we made an effort to develop CNN based method to detect distracted driver and identify the cause of distractions like talking, sleeping or eating by means of face and hand localization. Four architectures namely CNN, VGG-16, ResNet50 and MobileNetV2 have been adopted for transfer learning. To verify the effectiveness, the proposed model is trained with thousands of images from a publicly available dataset containing ten different postures or conditions of a distracted driver and analyzed the results using various performance metrics. The performance results showed that the pre-trained MobileNetV2 model has the best classification efficiency. © 2022 The Author(s)
- Authors: Hossain, Md Uzzol , Rahman, Md Ataur , Islam, Md Manowarul , Akhter, Arnisha , Uddin, Md Ashraf , Paul, Bikash
- Date: 2022
- Type: Text , Journal article
- Relation: Intelligent Systems with Applications Vol. 14, no. (2022), p.
- Full Text:
- Reviewed:
- Description: Recently, the number of road accidents has been increased worldwide due to the distraction of the drivers. This rapid road crush often leads to injuries, loss of properties, even deaths of the people. Therefore, it is essential to monitor and analyze the driver's behavior during the driving time to detect the distraction and mitigate the number of road accident. To detect various kinds of behavior like- using cell phone, talking to others, eating, sleeping or lack of concentration during driving; machine learning/deep learning can play significant role. However, this process may need high computational capacity to train the model by huge number of training dataset. In this paper, we made an effort to develop CNN based method to detect distracted driver and identify the cause of distractions like talking, sleeping or eating by means of face and hand localization. Four architectures namely CNN, VGG-16, ResNet50 and MobileNetV2 have been adopted for transfer learning. To verify the effectiveness, the proposed model is trained with thousands of images from a publicly available dataset containing ten different postures or conditions of a distracted driver and analyzed the results using various performance metrics. The performance results showed that the pre-trained MobileNetV2 model has the best classification efficiency. © 2022 The Author(s)
DINE : a framework for deep incomplete network embedding
- Hou, Ke, Liu, Jiaying, Peng, Yin, Xu, Bo, Lee, Ivan, Xia, Feng
- Authors: Hou, Ke , Liu, Jiaying , Peng, Yin , Xu, Bo , Lee, Ivan , Xia, Feng
- Date: 2019
- Type: Text , Conference paper
- Relation: 32nd Australasian Joint Conference on Artificial Intelligence, AI 2019 Vol. 11919 LNAI, p. 165-176
- Full Text:
- Reviewed:
- Description: Network representation learning (NRL) plays a vital role in a variety of tasks such as node classification and link prediction. It aims to learn low-dimensional vector representations for nodes based on network structures or node attributes. While embedding techniques on complete networks have been intensively studied, in real-world applications, it is still a challenging task to collect complete networks. To bridge the gap, in this paper, we propose a Deep Incomplete Network Embedding method, namely DINE. Specifically, we first complete the missing part including both nodes and edges in a partially observable network by using the expectation-maximization framework. To improve the embedding performance, we consider both network structures and node attributes to learn node representations. Empirically, we evaluate DINE over three networks on multi-label classification and link prediction tasks. The results demonstrate the superiority of our proposed approach compared against state-of-the-art baselines. © 2019, Springer Nature Switzerland AG.
- Description: E1
- Authors: Hou, Ke , Liu, Jiaying , Peng, Yin , Xu, Bo , Lee, Ivan , Xia, Feng
- Date: 2019
- Type: Text , Conference paper
- Relation: 32nd Australasian Joint Conference on Artificial Intelligence, AI 2019 Vol. 11919 LNAI, p. 165-176
- Full Text:
- Reviewed:
- Description: Network representation learning (NRL) plays a vital role in a variety of tasks such as node classification and link prediction. It aims to learn low-dimensional vector representations for nodes based on network structures or node attributes. While embedding techniques on complete networks have been intensively studied, in real-world applications, it is still a challenging task to collect complete networks. To bridge the gap, in this paper, we propose a Deep Incomplete Network Embedding method, namely DINE. Specifically, we first complete the missing part including both nodes and edges in a partially observable network by using the expectation-maximization framework. To improve the embedding performance, we consider both network structures and node attributes to learn node representations. Empirically, we evaluate DINE over three networks on multi-label classification and link prediction tasks. The results demonstrate the superiority of our proposed approach compared against state-of-the-art baselines. © 2019, Springer Nature Switzerland AG.
- Description: E1
AI and IoT-Enabled smart exoskeleton system for rehabilitation of paralyzed people in connected communities
- Jacob, Sunil, Alagirisamy, Mukil, Xi, Chen, Balasubramanian, Venki, Srinivasan, Ram
- Authors: Jacob, Sunil , Alagirisamy, Mukil , Xi, Chen , Balasubramanian, Venki , Srinivasan, Ram
- Date: 2021
- Type: Text , Journal article
- Relation: IEEE Access Vol. 9, no. (2021), p. 80340-80350
- Full Text:
- Reviewed:
- Description: In recent years, the number of cases of spinal cord injuries, stroke and other nervous impairments have led to an increase in the number of paralyzed patients worldwide. Rehabilitation that can aid and enhance the lives of such patients is the need of the hour. Exoskeletons have been found as one of the popular means of rehabilitation. The existing exoskeletons use techniques that impose limitations on adaptability, instant response and continuous control. Also most of them are expensive, bulky, and requires high level of training. To overcome all the above limitations, this paper introduces an Artificial Intelligence (AI) powered Smart and light weight Exoskeleton System (AI-IoT-SES) which receives data from various sensors, classifies them intelligently and generates the desired commands via Internet of Things (IoT) for rendering rehabilitation and support with the help of caretakers for paralyzed patients in smart and connected communities. In the proposed system, the signals collected from the exoskeleton sensors are processed using AI-assisted navigation module, and helps the caretakers in guiding, communicating and controlling the movements of the exoskeleton integrated to the patients. The navigation module uses AI and IoT enabled Simultaneous Localization and Mapping (SLAM). The casualties of a paralyzed person are reduced by commissioning the IoT platform to exchange data from the intelligent sensors with the remote location of the caretaker to monitor the real time movement and navigation of the exoskeleton. The automated exoskeleton detects and take decisions on navigation thereby improving the life conditions of such patients. The experimental results simulated using MATLAB shows that the proposed system is the ideal method for rendering rehabilitation and support for paralyzed patients in smart communities. © 2013 IEEE. **Please note that there are multiple authors for this article therefore only the name of the first 5 including Federation University Australia affiliate “Venki Balasubramanian” is provided in this record**
- Authors: Jacob, Sunil , Alagirisamy, Mukil , Xi, Chen , Balasubramanian, Venki , Srinivasan, Ram
- Date: 2021
- Type: Text , Journal article
- Relation: IEEE Access Vol. 9, no. (2021), p. 80340-80350
- Full Text:
- Reviewed:
- Description: In recent years, the number of cases of spinal cord injuries, stroke and other nervous impairments have led to an increase in the number of paralyzed patients worldwide. Rehabilitation that can aid and enhance the lives of such patients is the need of the hour. Exoskeletons have been found as one of the popular means of rehabilitation. The existing exoskeletons use techniques that impose limitations on adaptability, instant response and continuous control. Also most of them are expensive, bulky, and requires high level of training. To overcome all the above limitations, this paper introduces an Artificial Intelligence (AI) powered Smart and light weight Exoskeleton System (AI-IoT-SES) which receives data from various sensors, classifies them intelligently and generates the desired commands via Internet of Things (IoT) for rendering rehabilitation and support with the help of caretakers for paralyzed patients in smart and connected communities. In the proposed system, the signals collected from the exoskeleton sensors are processed using AI-assisted navigation module, and helps the caretakers in guiding, communicating and controlling the movements of the exoskeleton integrated to the patients. The navigation module uses AI and IoT enabled Simultaneous Localization and Mapping (SLAM). The casualties of a paralyzed person are reduced by commissioning the IoT platform to exchange data from the intelligent sensors with the remote location of the caretaker to monitor the real time movement and navigation of the exoskeleton. The automated exoskeleton detects and take decisions on navigation thereby improving the life conditions of such patients. The experimental results simulated using MATLAB shows that the proposed system is the ideal method for rendering rehabilitation and support for paralyzed patients in smart communities. © 2013 IEEE. **Please note that there are multiple authors for this article therefore only the name of the first 5 including Federation University Australia affiliate “Venki Balasubramanian” is provided in this record**
Melanoma classification using efficientnets and ensemble of models with different input resolution
- Karki, Sagar, Kulkarni, Pradnya, Stranieri, Andrew
- Authors: Karki, Sagar , Kulkarni, Pradnya , Stranieri, Andrew
- Date: 2021
- Type: Text , Conference paper
- Relation: 2021 Australasian Computer Science Week Multiconference, ACSW 2021, Virtual, Online, 1-5 February 2021, ACM International Conference Proceeding Series
- Full Text:
- Reviewed:
- Description: Early and accurate detection of melanoma with data analytics can make treatment more effective. This paper proposes a method to classify melanoma cases using deep learning on dermoscopic images. The method demonstrates that heavy augmentation during training and testing produces promising results and warrants further research. The proposed method has been evaluated on the SIIM-ISIC Melanoma Classification 2020 dataset and the best ensemble model achieved 0.9411 area under the ROC curve on hold out test data. © 2021 ACM.
- Authors: Karki, Sagar , Kulkarni, Pradnya , Stranieri, Andrew
- Date: 2021
- Type: Text , Conference paper
- Relation: 2021 Australasian Computer Science Week Multiconference, ACSW 2021, Virtual, Online, 1-5 February 2021, ACM International Conference Proceeding Series
- Full Text:
- Reviewed:
- Description: Early and accurate detection of melanoma with data analytics can make treatment more effective. This paper proposes a method to classify melanoma cases using deep learning on dermoscopic images. The method demonstrates that heavy augmentation during training and testing produces promising results and warrants further research. The proposed method has been evaluated on the SIIM-ISIC Melanoma Classification 2020 dataset and the best ensemble model achieved 0.9411 area under the ROC curve on hold out test data. © 2021 ACM.
Assessing trust level of a driverless car using deep learning
- Karmakar, Gour, Chowdhury, Abdullahi, Das, Rajkumar, Kamruzzaman, Joarder, Islam, Syed
- Authors: Karmakar, Gour , Chowdhury, Abdullahi , Das, Rajkumar , Kamruzzaman, Joarder , Islam, Syed
- Date: 2021
- Type: Text , Journal article
- Relation: IEEE Transactions on Intelligent Transportation Systems Vol. 22, no. 7 (2021), p. 4457-4466
- Full Text: false
- Reviewed:
- Description: The increasing adoption of driverless cars already providing a shift to move away from traditional transportation systems to automated ones in many industrial and commercial applications. Recent research has justified that driverless vehicles will considerably reduce traffic congestions, accidents, carbon emissions, and enhance the accessibility of driving to wider cross-section of people and lifestyle choices. However, at present, people's main concerns are about its privacy and security. Since traditional protocol layers based security mechanisms are not so effective for a distributed system, trust value-based security mechanisms, a type of pervasive security, are appearing as popular and promising techniques. A few statistical non-learning based models for measuring the trust level of a driverless are available in the current literature. These are not so effective because of not being able to capture the extremely distributed, dynamic, and complex nature of the traffic systems. To bridge this research gap, in this paper, for the first time, we propose two deep learning-based models that measure the trustworthiness of a driverless car and its major On-Board Unit (OBU) components. The second model also determines its OBU components that were breached during the driving operation. Results produced using real and simulated traffic data demonstrate that our proposed DNN based deep learning models outperform other machine learning models in assessing the trustworthiness of individual car as well as its OBU components. The average precision of detection accuracies for the car, LiDAR, camera, and radar are 0.99, 0.96, 0.81, and 0.83, respectively, which indicates the potential real-life application of our models in assessing the trust level of a driverless car. © 2000-2011 IEEE.
Deep learning : survey of environmental and camera impacts on internet of things images
- Kaur, Roopdeep, Karmakar, Gour, Xia, Feng, Imran, Muhammad
- Authors: Kaur, Roopdeep , Karmakar, Gour , Xia, Feng , Imran, Muhammad
- Date: 2023
- Type: Text , Journal article
- Relation: Artificial Intelligence Review Vol. 56, no. 9 (2023), p. 9605-9638
- Full Text:
- Reviewed:
- Description: Internet of Things (IoT) images are captivating growing attention because of their wide range of applications which requires visual analysis to drive automation. However, IoT images are predominantly captured from outdoor environments and thus are inherently impacted by the camera and environmental parameters which can adversely affect corresponding applications. Deep Learning (DL) has been widely adopted in the field of image processing and computer vision and can reduce the impact of these parameters on IoT images. Albeit, there are many DL-based techniques available in the current literature for analyzing and reducing the environmental and camera impacts on IoT images. However, to the best of our knowledge, no survey paper presents state-of-the-art DL-based approaches for this purpose. Motivated by this, for the first time, we present a Systematic Literature Review (SLR) of existing DL techniques available for analyzing and reducing environmental and camera lens impacts on IoT images. As part of this SLR, firstly, we reiterate and highlight the significance of IoT images in their respective applications. Secondly, we describe the DL techniques employed for assessing the environmental and camera lens distortion impacts on IoT images. Thirdly, we illustrate how DL can be effective in reducing the impact of environmental and camera lens distortion in IoT images. Finally, along with the critical reflection on the advantages and limitations of the techniques, we also present ways to address the research challenges of existing techniques and identify some further researches to advance the relevant research areas. © 2023, The Author(s).
- Authors: Kaur, Roopdeep , Karmakar, Gour , Xia, Feng , Imran, Muhammad
- Date: 2023
- Type: Text , Journal article
- Relation: Artificial Intelligence Review Vol. 56, no. 9 (2023), p. 9605-9638
- Full Text:
- Reviewed:
- Description: Internet of Things (IoT) images are captivating growing attention because of their wide range of applications which requires visual analysis to drive automation. However, IoT images are predominantly captured from outdoor environments and thus are inherently impacted by the camera and environmental parameters which can adversely affect corresponding applications. Deep Learning (DL) has been widely adopted in the field of image processing and computer vision and can reduce the impact of these parameters on IoT images. Albeit, there are many DL-based techniques available in the current literature for analyzing and reducing the environmental and camera impacts on IoT images. However, to the best of our knowledge, no survey paper presents state-of-the-art DL-based approaches for this purpose. Motivated by this, for the first time, we present a Systematic Literature Review (SLR) of existing DL techniques available for analyzing and reducing environmental and camera lens impacts on IoT images. As part of this SLR, firstly, we reiterate and highlight the significance of IoT images in their respective applications. Secondly, we describe the DL techniques employed for assessing the environmental and camera lens distortion impacts on IoT images. Thirdly, we illustrate how DL can be effective in reducing the impact of environmental and camera lens distortion in IoT images. Finally, along with the critical reflection on the advantages and limitations of the techniques, we also present ways to address the research challenges of existing techniques and identify some further researches to advance the relevant research areas. © 2023, The Author(s).
Impact of traditional and embedded image denoising on CNN-based deep learning
- Kaur, Roopdeep, Karmakar, Gour, Imran, Muhammad
- Authors: Kaur, Roopdeep , Karmakar, Gour , Imran, Muhammad
- Date: 2023
- Type: Text , Journal article
- Relation: Applied sciences Vol. 13, no. 20 (2023), p.
- Full Text:
- Reviewed:
- Description: In digital image processing, filtering noise is an important step for reconstructing a high-quality image for further processing such as object segmentation, object detection, and object recognition. Various image-denoising approaches, including median, Gaussian, and bilateral filters, are available in the literature. Since convolutional neural networks (CNN) are able to directly learn complex patterns and features from data, they have become a popular choice for image-denoising tasks. As a result of their ability to learn and adapt to various denoising scenarios, CNNs are powerful tools for image denoising. Some deep learning techniques such as CNN incorporate denoising strategies directly into the CNN model layers. A primary limitation of these methods is their necessity to resize images to a consistent size. This resizing can result in a loss of vital image details, which might compromise CNN’s effectiveness. Because of this issue, we utilize a traditional denoising method as a preliminary step for noise reduction before applying CNN. To our knowledge, a comparative performance study of CNN using traditional and embedded denoising against a baseline approach (without denoising) is yet to be performed. To analyze the impact of denoising on the CNN performance, in this paper, firstly, we filter the noise from the images using traditional means of denoising method before their use in the CNN model. Secondly, we embed a denoising layer in the CNN model. To validate the performance of image denoising, we performed extensive experiments for both traffic sign and object recognition datasets. To decide whether denoising will be adopted and to decide on the type of filter to be used, we also present an approach exploiting the peak-signal-to-noise-ratio (PSNRs) distribution of images. Both CNN accuracy and PSNRs distribution are used to evaluate the effectiveness of the denoising approaches. As expected, the results vary with the type of filter, impact, and dataset used in both traditional and embedded denoising approaches. However, traditional denoising shows better accuracy, while embedded denoising shows lower computational time for most of the cases. Overall, this comparative study gives insights into whether denoising will be adopted in various CNN-based image analyses, including autonomous driving, animal detection, and facial recognition.
- Authors: Kaur, Roopdeep , Karmakar, Gour , Imran, Muhammad
- Date: 2023
- Type: Text , Journal article
- Relation: Applied sciences Vol. 13, no. 20 (2023), p.
- Full Text:
- Reviewed:
- Description: In digital image processing, filtering noise is an important step for reconstructing a high-quality image for further processing such as object segmentation, object detection, and object recognition. Various image-denoising approaches, including median, Gaussian, and bilateral filters, are available in the literature. Since convolutional neural networks (CNN) are able to directly learn complex patterns and features from data, they have become a popular choice for image-denoising tasks. As a result of their ability to learn and adapt to various denoising scenarios, CNNs are powerful tools for image denoising. Some deep learning techniques such as CNN incorporate denoising strategies directly into the CNN model layers. A primary limitation of these methods is their necessity to resize images to a consistent size. This resizing can result in a loss of vital image details, which might compromise CNN’s effectiveness. Because of this issue, we utilize a traditional denoising method as a preliminary step for noise reduction before applying CNN. To our knowledge, a comparative performance study of CNN using traditional and embedded denoising against a baseline approach (without denoising) is yet to be performed. To analyze the impact of denoising on the CNN performance, in this paper, firstly, we filter the noise from the images using traditional means of denoising method before their use in the CNN model. Secondly, we embed a denoising layer in the CNN model. To validate the performance of image denoising, we performed extensive experiments for both traffic sign and object recognition datasets. To decide whether denoising will be adopted and to decide on the type of filter to be used, we also present an approach exploiting the peak-signal-to-noise-ratio (PSNRs) distribution of images. Both CNN accuracy and PSNRs distribution are used to evaluate the effectiveness of the denoising approaches. As expected, the results vary with the type of filter, impact, and dataset used in both traditional and embedded denoising approaches. However, traditional denoising shows better accuracy, while embedded denoising shows lower computational time for most of the cases. Overall, this comparative study gives insights into whether denoising will be adopted in various CNN-based image analyses, including autonomous driving, animal detection, and facial recognition.
Robust Mobile Malware Detection
- Authors: Khoda, Mahbub
- Date: 2020
- Type: Text , Thesis , PhD
- Full Text:
- Description: The increasing popularity and use of smartphones and hand-held devices have made them the most popular target for malware attackers. Researchers have proposed machine learning-based models to automatically detect malware attacks on these devices. Since these models learn application behaviors solely from the extracted features, choosing an appropriate and meaningful feature set is one of the most crucial steps for designing an effective mobile malware detection system. There are four categories of features for mobile applications. Previous works have taken arbitrary combinations of these categories to design models, resulting in sub-optimal performance. This thesis systematically investigates the individual impact of these feature categories on mobile malware detection systems. Feature categories that complement each other are investigated and categories that add redundancy to the feature space (thereby degrading the performance) are analyzed. In the process, the combination of feature categories that provides the best detection results is identified. Ensuring reliability and robustness of the above-mentioned malware detection systems is of utmost importance as newer techniques to break down such systems continue to surface. Adversarial attack is one such evasive attack that can bypass a detection system by carefully morphing a malicious sample even though the sample was originally correctly identified by the same system. Self-crafted adversarial samples can be used to retrain a model to defend against such attacks. However, randomly using too many such samples, as is currently done in the literature, can further degrade detection performance. This work proposed two intelligent approaches to retrain a classifier through the intelligent selection of adversarial samples. The first approach adopts a distance-based scheme where the samples are chosen based on their distance from malware and benign cluster centers while the second selects the samples based on a probability measure derived from a kernel-based learning method. The second method achieved a 6% improvement in terms of accuracy. To ensure practical deployment of malware detection systems, it is necessary to keep the real-world data characteristics in mind. For example, the benign applications deployed in the market greatly outnumber malware applications. However, most studies have assumed a balanced data distribution. Also, techniques to handle imbalanced data in other domains cannot be applied directly to mobile malware detection since they generate synthetic samples with broken functionality, making them invalid. In this regard, this thesis introduces a novel synthetic over-sampling technique that ensures valid sample generation. This technique is subsequently combined with a dynamic cost function in the learning scheme that automatically adjusts minority class weight during model training which counters the bias towards the majority class and stabilizes the model. This hybrid method provided a 9% improvement in terms of F1-score. Aiming to design a robust malware detection system, this thesis extensively studies machine learning-based mobile malware detection in terms of best feature category combination, resilience against evasive attacks, and practical deployment of detection models. Given the increasing technological advancements in mobile and hand-held devices, this study will be very useful for designing robust cybersecurity systems to ensure safe usage of these devices.
- Description: Doctor of Philosophy
- Authors: Khoda, Mahbub
- Date: 2020
- Type: Text , Thesis , PhD
- Full Text:
- Description: The increasing popularity and use of smartphones and hand-held devices have made them the most popular target for malware attackers. Researchers have proposed machine learning-based models to automatically detect malware attacks on these devices. Since these models learn application behaviors solely from the extracted features, choosing an appropriate and meaningful feature set is one of the most crucial steps for designing an effective mobile malware detection system. There are four categories of features for mobile applications. Previous works have taken arbitrary combinations of these categories to design models, resulting in sub-optimal performance. This thesis systematically investigates the individual impact of these feature categories on mobile malware detection systems. Feature categories that complement each other are investigated and categories that add redundancy to the feature space (thereby degrading the performance) are analyzed. In the process, the combination of feature categories that provides the best detection results is identified. Ensuring reliability and robustness of the above-mentioned malware detection systems is of utmost importance as newer techniques to break down such systems continue to surface. Adversarial attack is one such evasive attack that can bypass a detection system by carefully morphing a malicious sample even though the sample was originally correctly identified by the same system. Self-crafted adversarial samples can be used to retrain a model to defend against such attacks. However, randomly using too many such samples, as is currently done in the literature, can further degrade detection performance. This work proposed two intelligent approaches to retrain a classifier through the intelligent selection of adversarial samples. The first approach adopts a distance-based scheme where the samples are chosen based on their distance from malware and benign cluster centers while the second selects the samples based on a probability measure derived from a kernel-based learning method. The second method achieved a 6% improvement in terms of accuracy. To ensure practical deployment of malware detection systems, it is necessary to keep the real-world data characteristics in mind. For example, the benign applications deployed in the market greatly outnumber malware applications. However, most studies have assumed a balanced data distribution. Also, techniques to handle imbalanced data in other domains cannot be applied directly to mobile malware detection since they generate synthetic samples with broken functionality, making them invalid. In this regard, this thesis introduces a novel synthetic over-sampling technique that ensures valid sample generation. This technique is subsequently combined with a dynamic cost function in the learning scheme that automatically adjusts minority class weight during model training which counters the bias towards the majority class and stabilizes the model. This hybrid method provided a 9% improvement in terms of F1-score. Aiming to design a robust malware detection system, this thesis extensively studies machine learning-based mobile malware detection in terms of best feature category combination, resilience against evasive attacks, and practical deployment of detection models. Given the increasing technological advancements in mobile and hand-held devices, this study will be very useful for designing robust cybersecurity systems to ensure safe usage of these devices.
- Description: Doctor of Philosophy