Internet of Things (IoT) devices for the home have made a lot of people’s lives better, but their popularity has also raised privacy and safety concerns. This study explores the application of deep learning models for anomaly detection and face recognition in IoT devices within the context of smart homes. Six models, namely, LR-XGB-CNN, LR-GBC-CNN, LR-CBC-CNN, LR-HGBC-CNN, LR-ABC-CNN, and LR-LGBM-CNN, were proposed and evaluated for their performance. The models were trained and tested on labeled datasets of sensor readings and face images, using a range of performance metrics to assess their effectiveness. Performance evaluations were conducted for each of the proposed models, revealing their strengths and areas for improvement. Comparative analysis of the models showed that the LR-HGBC-CNN model consistently outperformed the others in both anomaly detection and face recognition tasks, achieving high accuracy, precision, recall, F1 score, and AUC-ROC values. For anomaly detection, the LR-HGBC-CNN model achieved an accuracy of 94%, a precision of 91%, a recall of 96%, an F1 score of 93%, and an AUC-ROC of 0.96. In face recognition, the LR-HGBC-CNN model demonstrated an accuracy of 88%, precision of 86%, recall of 90%, F1 score of 88%, and an AUC-ROC of 0.92. The models exhibited promising capabilities in detecting anomalies, recognizing faces, and integrating these functionalities within smart home IoT devices. The study’s findings underscore the potential of deep learning approaches for enhancing security and privacy in smart homes. However, further research is warranted to evaluate the models’ generalizability, explore advanced techniques such as transfer learning and hybrid methods, investigate privacy-preserving mechanisms, and address deployment challenges.

Recurrent Neural Networks (RNNs) and their variants have been demonstrated tremendous successes in modeling sequential data such as audio processing, video processing, time series analysis, and text mining. Inspired by these facts, we propose human activity recognition technique to proceed visual data via utilizing convolution neural network (CNN) and Bidirectional-gated recurrent unit (Bi-GRU). Firstly, we extract deep features from frames sequence of human activities videos using CNN and then select most important features from the deep appearances to improve performance and decrease computational complexity of the model. Secondly, to learn temporal motions of frames sequence, we design Bi-GRU and feed those deep-important features extracted from frames sequence of human activities to Bi-GRU which learn temporal dynamics in forward and backward direction at each time step. We conduct extensive experiments on realistic videos of human activity recognition datasets YouTube11, HMDB51 and UCF101. Lastly, we compare the obtained results with existing methods to show the competence of our proposed technique.

Smart video surveillance plays a significant role in public security via storing a huge amount of continual stream data, evaluates them, and generates warns where undesirable human activities are performed. Recognition of human activities in video surveillance has faced many challenges such as optimal evaluation of human activities under growing volumes of streaming data with complex computation and huge time processing complexity. To tackle these challenges we introduce a lightweighted spatial-deep features integration using multilayer GRU (SDIGRU). First, we extract spatial and deep features from frames sequence of realistic human activity videos via utilizing a lightweight MobileNetV2 model and then integrate those spatial-deep features. Although deep features can be used for human activity recognition, they contain only the high-level appearance, which is insufficient to correctly identify the particular activity of human. Thus, we jointly apply deep information with spatial appearance to produce detailed level information. Furthermore, we select rich informative features from spatial-deep appearances. Then, we train multilayer gated recurrent unit (GRU) and feed informative features to learn the temporal dynamics of human activity frames sequence at each time step of GRU. We conduct our experiments on benchmark YouTube11, HMDB51, and UCF101 datasets of human activity recognition. The empirical results show that our method achieved significant recognition performance with low computational complexity and quick response. Finally, we compare the results with existing state-of-the-art techniques, which show the effectiveness of our method.

With the proliferation of connected devices in the Internet of Things (IoT), cybersecurity threats have increased. Identifying malicious attacks in IoT requires advanced techniques tailored to this ecosystem. Existing algorithms have limited effectiveness in detecting obfuscated IoT malware. This study proposes the Elucidating Cybersecurity-promulgated Malware Taxonomy (ECMT) framework, combining memory analysis and ensemble machine learning, to enhance IoT malware categorization. ECMT integrates Support Vector Classification, Quadratic Discriminant Analysis, and AdaBoost on forensic artifacts from memory dumps to improve detection across families like ransomware, spyware, and trojans. ECMT can enable intrusion prevention, information protection, and cybercrime deterrence in IoT environments. Experiments on a balanced dataset indicate AdaBoost achieved 96% accuracy, demonstrating ECMT’s capabilities against complex IoT threats. The integrated approach provides automated, adaptable detection scalable to large IoT deployments through efficient linear models and robust ensemble learning. ECMT addresses concept drift and interpretability via retraining and explanation techniques. Results highlight advanced memory analysis and optimized machine learning classifiers as a promising solution for robust IoT malware detection despite adversaries’ evolving tactics. Further research can extend platform support, harden models against attacks, and refine streaming input. ECMT establishes a foundation for IoT security by unifying memory forensics, optimized neural architectures, and tailored ensemble learning.

Authorship verification is a crucial task in digital forensic investigations, where it is often necessary to determine whether a specific individual wrote a particular piece of text. Convolutional Neural Networks (CNNs) have shown promise in solving this problem, but their performance highly depends on the choice of hyperparameters. In this paper, we explore the effectiveness of hyperparameter tuning in improving the performance of CNNs for authorship verification. We conduct experiments using a Hyper Tuned CNN model with three popular optimization algorithms: Adaptive Moment Estimation (ADAM), Stochastic Gradient Descent (SGD), and Root Mean Squared Propagation (RMSPROP). The model is trained and tested on a dataset of text samples collected from various authors, and the performance is evaluated using accuracy, precision, recall, and F1 score. We compare the performance of the three optimization algorithms and demonstrate the effectiveness of hyperparameter tuning in improving the accuracy of the CNN model. Our results show that the Hyper Tuned CNN model with ADAM Optimizer achieves the highest accuracy of up to 90%. Furthermore, we demonstrate that hyperparameter tuning can help achieve significant performance improvements, even using a relatively simple model architecture like CNNs. Our findings suggest that the choice of the optimization algorithm is a crucial factor in the performance of CNNs for authorship verification and that hyperparameter tuning can be an effective way to optimize this choice. Overall, this paper demonstrates the effectiveness of hyperparameter tuning in improving the performance of CNNs for authorship verification in digital forensic investigations. Our findings have important implications for developing accurate and reliable authorship verification systems, which are crucial for various applications in digital forensics, such as identifying the author of anonymous threatening messages or detecting cases of plagiarism.

Decision making system in smart healthcare of Heart failure (HF) disease can facilitate patients remotely, which is an active research area in medical fields, industrial applications, and academia. However, implementation of HF decision making systems on the edge using consumer electronic devices face challenges in processing huge suboptimal heart disease data due to limited computational capabilities. To alleviate such challenges, we introduce an optimize method of HF prediction using four statistical parameters such as strong correlation coefficient, mutual information, analysis of variance (ANOVA) f-test and chi-squared χ2. These parameters select best features through maximum likelihood based threshold value. We then use mean value fusion approach to construct a combine feature space. Afterwards, combine features further optimized via sequential forward selection (SFS) algorithm, which provide best subset of variables. Finally, a novel deep Multi-tier long short-term memory (LSTM) trained on top of the acquired features for diagnosis of HF. We conduct extensive experimentation on Cleveland heart disease dataset, which initially consist of 13 features. Moreover, after execution of our proposed optimize strategy, the size of features restricted to 8. Consequently, these optimal features provide enormous help in diagnosis procedure of HF. The competence of our propose technique evaluated using various metrics such as accuracy, sensitivity, precision, specificity, F1-score, Mathew correlation coefficient (MCC), receiver operating characteristic (ROC) and area under curve (AUC). As shown by the experimental results, the proposed features optimization and multi-tier LSTM strategy improve accuracy of 8.9% compared with different traditional classifiers.