Pub Date : 2022-06-06DOI: 10.1109/ISDFS55398.2022.9800823
Ray Kresman, L. Dunning, Jing-yan Lu
Privacy is an integral part of today’s digital transformation era and there is a paradigm shift toward anonymity in communication. This paper proposes an algorithm for assigning anonymous IDs to every member of a group of participants and provides an upper bound on one of the parameters of the algorithm. The algorithm terminates in one round.The proposed algorithm improves an earlier algorithm (AIDA) and mirrors its working. Simulation is used to show that AIDA may not terminate when the number of participants is large. We then discuss the software architecture and performance improvements offered by the proposed algorithm.
{"title":"An Improved Anonymous Identifier","authors":"Ray Kresman, L. Dunning, Jing-yan Lu","doi":"10.1109/ISDFS55398.2022.9800823","DOIUrl":"https://doi.org/10.1109/ISDFS55398.2022.9800823","url":null,"abstract":"Privacy is an integral part of today’s digital transformation era and there is a paradigm shift toward anonymity in communication. This paper proposes an algorithm for assigning anonymous IDs to every member of a group of participants and provides an upper bound on one of the parameters of the algorithm. The algorithm terminates in one round.The proposed algorithm improves an earlier algorithm (AIDA) and mirrors its working. Simulation is used to show that AIDA may not terminate when the number of participants is large. We then discuss the software architecture and performance improvements offered by the proposed algorithm.","PeriodicalId":114335,"journal":{"name":"2022 10th International Symposium on Digital Forensics and Security (ISDFS)","volume":"9 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-06-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"115411561","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-06-06DOI: 10.1109/ISDFS55398.2022.9800820
Mena Youssef, Serkan Varol, Serkan Catma
Motor vehicle crashes in the United States are one of the significant causes of death. Females involved in motor vehicle fatal accidents show a significant increase in the last decade. This project investigates variables captured within the National Highway Traffic Safety Administration’s reporting system to see the contributing factors toward female driver involvement in crashes that result in a fatality in Tennessee. The findings showed that variables such as driver height, weight, age, and vehicle’s model year have the most influence per mean decrease Gini on female involvement in accidents resulting in fatalities. Government officials can use evidence gained from this study to introduce laws and safety measures to help decrease the rate of fatal accidents.
{"title":"Vehicle Fatality Analysis by Gender using Predictive Analytics","authors":"Mena Youssef, Serkan Varol, Serkan Catma","doi":"10.1109/ISDFS55398.2022.9800820","DOIUrl":"https://doi.org/10.1109/ISDFS55398.2022.9800820","url":null,"abstract":"Motor vehicle crashes in the United States are one of the significant causes of death. Females involved in motor vehicle fatal accidents show a significant increase in the last decade. This project investigates variables captured within the National Highway Traffic Safety Administration’s reporting system to see the contributing factors toward female driver involvement in crashes that result in a fatality in Tennessee. The findings showed that variables such as driver height, weight, age, and vehicle’s model year have the most influence per mean decrease Gini on female involvement in accidents resulting in fatalities. Government officials can use evidence gained from this study to introduce laws and safety measures to help decrease the rate of fatal accidents.","PeriodicalId":114335,"journal":{"name":"2022 10th International Symposium on Digital Forensics and Security (ISDFS)","volume":"122 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-06-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"123042328","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-06-06DOI: 10.1109/ISDFS55398.2022.9800780
P. Charan, P. Anand, S. Shukla, N. Selvan, Hrushikesh Chunduri
Target specific malware is one of the major concerns for many global IT firms and government organizations. In recent times, state-sponsored Advanced Persistent Threat (APT) groups have evolved in developing more intelligent and targeted malware by misusing various legitimate services. This work sheds light on modeling a threat scenario to emphasize how targeted attacks are performed by misusing legitimate services (Google Teachable Machine in our scenario) for malicious purposes in establishing foothold, lateral movement, and data exfiltration phases of APT life cycle. As a proof of concept, we validate our threat model with five different experiments highlighting how an attacker can execute a personalized boot sector ransomware and fileless malware on a targeted individual in corporate networks. Furthermore, assuming the attacker has limited information regarding the target, we use sinGAN to generate synthetic image samples to train a model for identifying the targets. In addition, we present a correlation study between target prediction confidence and effective payload deployment for all experiments. In our observation, targeted file-less malware turned out to be quicker and pestilent, averaging 25.11 seconds to encrypt the whole disk with 80% target prediction confidence.
{"title":"DOTMUG: A Threat Model for Target Specific APT Attacks–Misusing Google Teachable Machine","authors":"P. Charan, P. Anand, S. Shukla, N. Selvan, Hrushikesh Chunduri","doi":"10.1109/ISDFS55398.2022.9800780","DOIUrl":"https://doi.org/10.1109/ISDFS55398.2022.9800780","url":null,"abstract":"Target specific malware is one of the major concerns for many global IT firms and government organizations. In recent times, state-sponsored Advanced Persistent Threat (APT) groups have evolved in developing more intelligent and targeted malware by misusing various legitimate services. This work sheds light on modeling a threat scenario to emphasize how targeted attacks are performed by misusing legitimate services (Google Teachable Machine in our scenario) for malicious purposes in establishing foothold, lateral movement, and data exfiltration phases of APT life cycle. As a proof of concept, we validate our threat model with five different experiments highlighting how an attacker can execute a personalized boot sector ransomware and fileless malware on a targeted individual in corporate networks. Furthermore, assuming the attacker has limited information regarding the target, we use sinGAN to generate synthetic image samples to train a model for identifying the targets. In addition, we present a correlation study between target prediction confidence and effective payload deployment for all experiments. In our observation, targeted file-less malware turned out to be quicker and pestilent, averaging 25.11 seconds to encrypt the whole disk with 80% target prediction confidence.","PeriodicalId":114335,"journal":{"name":"2022 10th International Symposium on Digital Forensics and Security (ISDFS)","volume":"37 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-06-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"125887562","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-06-06DOI: 10.1109/ISDFS55398.2022.9800790
A. Editya, T. Ahmad, H. Studiawan
Nowadays drones have an important role to help people in several aspects. For example, drones are very beneficial in military sector. In this sector, drones may have a collision with different reasons, such as human error and a malfunctioning system. Therefore, forensic investigation helps the authority to find out drone collision cause. One indication before having collision is a change of direction of the drone. This cause can be detected by analyzing drone flying direction which changes irregularly. In this paper, we propose to use direction estimation method to assist the forensic investigation of a drone collision. We apply optical flow methods, specifically Lucas-Kanade, Horn-Schunck, and Gunnar-Farnerback. Based on the experimental results, Lucas-Kanade technique can achieve the best direction estimation providing a specific motion vector and also a filter to reduce noise.
{"title":"Direction Estimation of Drone Collision Using Optical Flow for Forensic Investigation","authors":"A. Editya, T. Ahmad, H. Studiawan","doi":"10.1109/ISDFS55398.2022.9800790","DOIUrl":"https://doi.org/10.1109/ISDFS55398.2022.9800790","url":null,"abstract":"Nowadays drones have an important role to help people in several aspects. For example, drones are very beneficial in military sector. In this sector, drones may have a collision with different reasons, such as human error and a malfunctioning system. Therefore, forensic investigation helps the authority to find out drone collision cause. One indication before having collision is a change of direction of the drone. This cause can be detected by analyzing drone flying direction which changes irregularly. In this paper, we propose to use direction estimation method to assist the forensic investigation of a drone collision. We apply optical flow methods, specifically Lucas-Kanade, Horn-Schunck, and Gunnar-Farnerback. Based on the experimental results, Lucas-Kanade technique can achieve the best direction estimation providing a specific motion vector and also a filter to reduce noise.","PeriodicalId":114335,"journal":{"name":"2022 10th International Symposium on Digital Forensics and Security (ISDFS)","volume":"29 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-06-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"115052348","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-06-06DOI: 10.1109/ISDFS55398.2022.9800798
Yuri G. Gordienko, Oleksandr Rokovyi, Oleg Alienin, S. Stirenko
The problem of object detection by YOLOv4 deep neural network (DNN) is considered on Stanford drone dataset (SDD) with object classes (pedestrians, bicyclists, cars, skateboarders, golf carts, and buses) collected by Unmanned Aerial Vehicle (UAV) video surveillance. Some frames (images) with labels were extracted from videos of this dataset and structured in the open-access SDD frames (SDDF) version (https://www.kaggle.com/yoctoman/stanford-drone-dataset-frames). The context-aware data augmentation (CADA) was proposed to change bounding box (BB) sizes by some percentage of its width and height. To investigate the possible effect of the dataset labeling quality the "dirty" and "clean" dataset versions were prepared, which differ by the evaluation subset only. CADA procedures lead to significant improvement of performance by loss and mean average precision (mAP) that can be observed both for "dirty" and "clean" evaluation subsets in comparison to experiments without CADA. Moreover, CADA procedures allow to get the mAP values on the "dirty" (real) evaluation subset that can be similar (and for some classes higher even) to the mAP values on the "clean" (ground-truth - GT) evaluation subset without CADA procedures. This effect can be explained by increase of signal-to-noise ratios for object-to-background pairs after IN-like cropping CADA procedures and then by increase of variability of object-to-background pair after subsequent OUT-like enlarging CADA procedures. It should be noted the non-commutative nature of CADA-based retraining procedures because their reverse direction like first-OUT-then-IN CADA in contrast to first-IN-then-OUT CADA did not lead to such a big increase of mAP values. Several CADA-sequences were analyzed and the best strategy consists in first-IN-then-OUT CADA procedures, where the extent of decrease and increase of BBs width and height can be different for various applications and datasets.
{"title":"Context-Aware Data Augmentation for Efficient Object Detection by UAV Surveillance","authors":"Yuri G. Gordienko, Oleksandr Rokovyi, Oleg Alienin, S. Stirenko","doi":"10.1109/ISDFS55398.2022.9800798","DOIUrl":"https://doi.org/10.1109/ISDFS55398.2022.9800798","url":null,"abstract":"The problem of object detection by YOLOv4 deep neural network (DNN) is considered on Stanford drone dataset (SDD) with object classes (pedestrians, bicyclists, cars, skateboarders, golf carts, and buses) collected by Unmanned Aerial Vehicle (UAV) video surveillance. Some frames (images) with labels were extracted from videos of this dataset and structured in the open-access SDD frames (SDDF) version (https://www.kaggle.com/yoctoman/stanford-drone-dataset-frames). The context-aware data augmentation (CADA) was proposed to change bounding box (BB) sizes by some percentage of its width and height. To investigate the possible effect of the dataset labeling quality the \"dirty\" and \"clean\" dataset versions were prepared, which differ by the evaluation subset only. CADA procedures lead to significant improvement of performance by loss and mean average precision (mAP) that can be observed both for \"dirty\" and \"clean\" evaluation subsets in comparison to experiments without CADA. Moreover, CADA procedures allow to get the mAP values on the \"dirty\" (real) evaluation subset that can be similar (and for some classes higher even) to the mAP values on the \"clean\" (ground-truth - GT) evaluation subset without CADA procedures. This effect can be explained by increase of signal-to-noise ratios for object-to-background pairs after IN-like cropping CADA procedures and then by increase of variability of object-to-background pair after subsequent OUT-like enlarging CADA procedures. It should be noted the non-commutative nature of CADA-based retraining procedures because their reverse direction like first-OUT-then-IN CADA in contrast to first-IN-then-OUT CADA did not lead to such a big increase of mAP values. Several CADA-sequences were analyzed and the best strategy consists in first-IN-then-OUT CADA procedures, where the extent of decrease and increase of BBs width and height can be different for various applications and datasets.","PeriodicalId":114335,"journal":{"name":"2022 10th International Symposium on Digital Forensics and Security (ISDFS)","volume":"43 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-06-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"123979562","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-06-06DOI: 10.1109/ISDFS55398.2022.9800781
João Gonçalo Pereira, Joaquim Gonçalves
Human activity recognition (HAR) is important in people’s daily life, helping in both human-to-human interaction and interpersonal relations. In HAR, many studies are presented to show the best data and the best methods in order to predict activities with the most accuracy possible. These studies have different approaches to the problems that HAR present when the real-time is important. In this paper we aim to present some of the methods that exist as well as some of the existing dataset’s and understand the different techniques used. The results show that the CNN’s algorithms has better performance than the others, however more work need to be developed namely in production of adequate dataset’s for training
{"title":"Human Activity Recognition: A review","authors":"João Gonçalo Pereira, Joaquim Gonçalves","doi":"10.1109/ISDFS55398.2022.9800781","DOIUrl":"https://doi.org/10.1109/ISDFS55398.2022.9800781","url":null,"abstract":"Human activity recognition (HAR) is important in people’s daily life, helping in both human-to-human interaction and interpersonal relations. In HAR, many studies are presented to show the best data and the best methods in order to predict activities with the most accuracy possible. These studies have different approaches to the problems that HAR present when the real-time is important. In this paper we aim to present some of the methods that exist as well as some of the existing dataset’s and understand the different techniques used. The results show that the CNN’s algorithms has better performance than the others, however more work need to be developed namely in production of adequate dataset’s for training","PeriodicalId":114335,"journal":{"name":"2022 10th International Symposium on Digital Forensics and Security (ISDFS)","volume":"8 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-06-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"114374582","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Since YouTube became one of the sources of income, the number of users has increased significantly and the number of spammers who aim to spread viruses or to promote their videos and channels. These behaviors have led many YouTubers to close their channels or to disable the comments because YouTube does not have enough tools to prevent it. Filtering Arabic spam comments is a big challenge at all according to various dialects which hold a huge number of synonyms. In this work, we have classified these comments using different algorithms such as Decision Tree(DT), Support Vector Machine (SVM), Naive Bayes(NB), Random Forest, and k-Nearest Neighbor (k-NN).
{"title":"Detecting Arabic YouTube Spam Using Data Mining Techniques","authors":"Yahya M. Tashtoush, Areen Magableh, Omar Darwish, Lujain Smadi, Omar Alomari, Anood ALghazoo","doi":"10.1109/ISDFS55398.2022.9800840","DOIUrl":"https://doi.org/10.1109/ISDFS55398.2022.9800840","url":null,"abstract":"Since YouTube became one of the sources of income, the number of users has increased significantly and the number of spammers who aim to spread viruses or to promote their videos and channels. These behaviors have led many YouTubers to close their channels or to disable the comments because YouTube does not have enough tools to prevent it. Filtering Arabic spam comments is a big challenge at all according to various dialects which hold a huge number of synonyms. In this work, we have classified these comments using different algorithms such as Decision Tree(DT), Support Vector Machine (SVM), Naive Bayes(NB), Random Forest, and k-Nearest Neighbor (k-NN).","PeriodicalId":114335,"journal":{"name":"2022 10th International Symposium on Digital Forensics and Security (ISDFS)","volume":"27 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-06-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"131127049","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-06-06DOI: 10.1109/ISDFS55398.2022.9800777
S. Reddy, Santanu Mandal, Varanasi L. V. S. K. B. Kasyap, K. AswathyR.
The usage of social media has expanded in recent years, allowing them to get news from around the world at any time. This in turn, is questioning the authenticity of the news that is being spread both globally and locally. Fake news such as misinformation, gossips is widely disseminated on social media having a negative impact on society and lives of the people. As a result, much study is being is carried out in order to detect them. The data can be clustered into smaller groups based on the type of news using a few learning approaches. A novel method has been proposed for prediction of the authenticity of the news of the LIAR dataset [1] using Logistic Regression and a boosting algorithm eXtreme Gradient Boosting (XGBoost) for efficacy, computational pace and performance of the model. This method detects fake news by analyzing the semantic and syntactic connections between sentences. Various graphs (like heat maps, bar charts) are plotted to show the distribution of the authenticity of news and also to compare the predicted result with the actual one. The proposed strategy addresses the effects of the hoax's global spread. People are hungry for information to defend themselves and others in a community where humans are confronting large-scale risks from harms. Some key traits such as Sentimental features, Content-based features, Frequency features, and Hybrid features (combinations of two or more features) are incorporated for early prediction of fake news spread via social media. The liar dataset is used to train the method and tested for accurate results. The experimental accuracy is found out to be 98%.
{"title":"A Novel Approach to Detect Fake News Using eXtreme Gradient Boosting","authors":"S. Reddy, Santanu Mandal, Varanasi L. V. S. K. B. Kasyap, K. AswathyR.","doi":"10.1109/ISDFS55398.2022.9800777","DOIUrl":"https://doi.org/10.1109/ISDFS55398.2022.9800777","url":null,"abstract":"The usage of social media has expanded in recent years, allowing them to get news from around the world at any time. This in turn, is questioning the authenticity of the news that is being spread both globally and locally. Fake news such as misinformation, gossips is widely disseminated on social media having a negative impact on society and lives of the people. As a result, much study is being is carried out in order to detect them. The data can be clustered into smaller groups based on the type of news using a few learning approaches. A novel method has been proposed for prediction of the authenticity of the news of the LIAR dataset [1] using Logistic Regression and a boosting algorithm eXtreme Gradient Boosting (XGBoost) for efficacy, computational pace and performance of the model. This method detects fake news by analyzing the semantic and syntactic connections between sentences. Various graphs (like heat maps, bar charts) are plotted to show the distribution of the authenticity of news and also to compare the predicted result with the actual one. The proposed strategy addresses the effects of the hoax's global spread. People are hungry for information to defend themselves and others in a community where humans are confronting large-scale risks from harms. Some key traits such as Sentimental features, Content-based features, Frequency features, and Hybrid features (combinations of two or more features) are incorporated for early prediction of fake news spread via social media. The liar dataset is used to train the method and tested for accurate results. The experimental accuracy is found out to be 98%.","PeriodicalId":114335,"journal":{"name":"2022 10th International Symposium on Digital Forensics and Security (ISDFS)","volume":"52 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-06-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"132282710","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-06-06DOI: 10.1109/ISDFS55398.2022.9800796
Sai Dheeraj Gummadi, Anirban Ghosh, Yeswanth Vootla
A transfer learning-based convolutional neural network (CNN) architecture is used in the current study to differentiate parasitic malaria cell images from the healthy ones and localize the parasites in infected images using global average pooling(GAP) and heat map. Malaria is a serious malady that can even lead to death in the absence of timely diagnosis. With the use of computerized malaria diagnosis, the suggested solution tackles the problem of timely detection and eases the strain on health care. Three transfer learning-based neural network architectures are studied and compared in terms of their accuracy, precision, sensitivity and specificity. The optimal model with less number of false negatives was then interfaced with a newly developed web service which can be easily accessed and used by common people. The studied models were trained and evaluated on 27,558 single cell images, yielding a maximum accuracy of 96.88%, with 97.35% sensitivity, 96.41% specificity, 96.89% F1-Score, and 96.44% precision.
{"title":"Transfer Learning based Classification of Plasmodium Falciparum Parasitic Blood Smear Images","authors":"Sai Dheeraj Gummadi, Anirban Ghosh, Yeswanth Vootla","doi":"10.1109/ISDFS55398.2022.9800796","DOIUrl":"https://doi.org/10.1109/ISDFS55398.2022.9800796","url":null,"abstract":"A transfer learning-based convolutional neural network (CNN) architecture is used in the current study to differentiate parasitic malaria cell images from the healthy ones and localize the parasites in infected images using global average pooling(GAP) and heat map. Malaria is a serious malady that can even lead to death in the absence of timely diagnosis. With the use of computerized malaria diagnosis, the suggested solution tackles the problem of timely detection and eases the strain on health care. Three transfer learning-based neural network architectures are studied and compared in terms of their accuracy, precision, sensitivity and specificity. The optimal model with less number of false negatives was then interfaced with a newly developed web service which can be easily accessed and used by common people. The studied models were trained and evaluated on 27,558 single cell images, yielding a maximum accuracy of 96.88%, with 97.35% sensitivity, 96.41% specificity, 96.89% F1-Score, and 96.44% precision.","PeriodicalId":114335,"journal":{"name":"2022 10th International Symposium on Digital Forensics and Security (ISDFS)","volume":"4 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-06-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"128818043","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-06-06DOI: 10.1109/ISDFS55398.2022.9800802
Mohd. Maaz Khan, Irfan Mabood Khan, Omar Farooq
Epilepsy is a severe neurological disorder, which is onset by the abrupt and erratic electrical gushing in the neurons. Epileptic seizures can be diagnosed by monitoring the brain’s electrical activity using Electroencephalogram (EEG) signals. Conventionally this analysis was done manually by neurologists and had various limitations, but now it is increasingly being automated to save time, minimize human errors and relieve the neurologists from excessive burden. In this study, the EEG signals are first converted into spectrograms. These spectrograms are then fed into the proposed Convolutional Neural Network (CNN) model to automatically learn the robust features and perform binary classification. The proposed CNN model, containing only 3.94 million parameters, obtained an accuracy of 90.9% and achieved precision, recall, and AUC of 91.1%, 93.5% and 97.9% respectively. This work is extended by applying transfer learning on four pre-trained networks VGG16, ResNet, DenseNet, and Inception using the same dataset. Among all these networks, DenseNet achieves the best performance having an accuracy of 92.6% followed by ResNet with an accuracy of 90.3%, Inception with an accuracy of 88.8%, and VGG16 having an accuracy of 88.5%. Although DenseNet achieves slightly better accuracy than the proposed CNN model, it contains almost twice the parameters (8.1 million) in the base model.
{"title":"Deep Learning based Lightweight Model for Seizure Detection using Spectrogram Images","authors":"Mohd. Maaz Khan, Irfan Mabood Khan, Omar Farooq","doi":"10.1109/ISDFS55398.2022.9800802","DOIUrl":"https://doi.org/10.1109/ISDFS55398.2022.9800802","url":null,"abstract":"Epilepsy is a severe neurological disorder, which is onset by the abrupt and erratic electrical gushing in the neurons. Epileptic seizures can be diagnosed by monitoring the brain’s electrical activity using Electroencephalogram (EEG) signals. Conventionally this analysis was done manually by neurologists and had various limitations, but now it is increasingly being automated to save time, minimize human errors and relieve the neurologists from excessive burden. In this study, the EEG signals are first converted into spectrograms. These spectrograms are then fed into the proposed Convolutional Neural Network (CNN) model to automatically learn the robust features and perform binary classification. The proposed CNN model, containing only 3.94 million parameters, obtained an accuracy of 90.9% and achieved precision, recall, and AUC of 91.1%, 93.5% and 97.9% respectively. This work is extended by applying transfer learning on four pre-trained networks VGG16, ResNet, DenseNet, and Inception using the same dataset. Among all these networks, DenseNet achieves the best performance having an accuracy of 92.6% followed by ResNet with an accuracy of 90.3%, Inception with an accuracy of 88.8%, and VGG16 having an accuracy of 88.5%. Although DenseNet achieves slightly better accuracy than the proposed CNN model, it contains almost twice the parameters (8.1 million) in the base model.","PeriodicalId":114335,"journal":{"name":"2022 10th International Symposium on Digital Forensics and Security (ISDFS)","volume":"95 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-06-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"124299602","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：