Saheed Ademola Bello

With the rapid growth of internet usage and the increasing number of connected devices, there is a critical need for advanced Network Traffic Classification (NTC) solutions to ensure optimal performance and robust security. Traditional NTC methods, such as port-based analysis and deep packet inspection, struggle to cope with modern network complexities, particularly dynamic port allocation and encrypted traffic. Recently, Convolutional Neural Networks (CNN) and Recurrent Neural Networks (RNN) have been employed to develop classification models to accomplish this task. Existing models for NTC often require significant computational resources due to their large number of parameters, leading to slower inference times and higher memory consumption. To overcome these limitations, we introduce a lightweight NTC model based on Depthwise Separable Convolutions and compare its performance against CNN, RNN, and state-of-the-art models. In terms of computational efficiency, our proposed lightweight CNN exhibits a markedly reduced computational footprint. It utilizes only 30,611 parameters and 0.627 MFLOPS, achieving inference times of 1.49 seconds on the CPU and 0.43 seconds on the GPU. This corresponds to roughly 4× fewer FLOPS than the RNN baseline and 16× fewer than the CNN baseline, while also offering an ultracompact design compared to state-of-the-art models. Such efficiency makes it exceptionally well-suited for real-time applications in resource-constrained environments. In addition, we have integrated eXplainable Artificial Intelligence techniques, specifically LIME and SHAP, to provide valuable insights into model predictions. LIME and SHAP help interpret the contribution of each feature in decision-making, enhancing the transparency and trust in the model's predictions, without compromising its lightweight nature. To support reproducibility and foster collaborative development, all associated code and resources have been made publicly available.

Cancer remains a major health threat with rising incidence and mortality rates. Despite the efficacy of chemotherapy, its lack of selectivity and associated severe side effects highlight the need for new, targeted anticancer therapies. Anticancer peptides (ACPs) have emerged as a promising alternative due to their biocompatibility, broad-spectrum anticancer activity, and unique mechanisms of action. This study presents a novel computational approach to design and identify ACPs using a multi-tier filtration system. Our method begins with peptide sequence generation via a recurrent neural network (RNN) trained on the acp740 dataset. The generated sequences undergo rigorous filtration: Tier-1 employs three deep learning-based classifiers (ACP-DL, ACP-MHCNN, ACP-LSE) to identify potential ACPs; Tier-2 uses a nearest centroid classifier to filter out statistically less relevant sequences; Tier-3 involves a final filtration using unsupervised nearest neighbor learning based on fused feature encoding schemes (CKSAAP, k-Mer, and BPF). Experimental results demonstrate a significant improvement in identifying viable ACP candidates, with the proposed method showing a 2.21-fold higher hit-rate compared to random sequence generation. Further analysis using t-SNE, PCA, and antimicrobial peptide (AMP) prediction tools confirms the robustness and effectiveness of the selected ACPs. Furthermore, performance comparisons using the proposed sequence filtering technique reveal that it surpasses the baseline LSTM and RNN-based sequence generation models by 2.95% and 14.11%, respectively. Complementary reverse analyses further validate the robustness and effectiveness of proposed sequence generation framework. The proposed computational approach offers a streamlined and economical alternative to traditional experimental methods, expediting the discovery of new ACPs and enhancing the accuracy of anticancer peptide predictions. The relevant models, codes, and results are also available on the authors github page at (https://github.com/mhdshl/ACP-Seq2Seq).

For energy management and billing purposes, advanced metering infrastructure (AMI) requires periodic transmission of consumer power consumption readings by smart meters to the electric utility (EU). An efficient way for collecting readings is the change-and-transmit approach (CAT AMI) whereby readings are only transmitted if there is an enough change in consumption readings. CAT AMI, however, is plagued by malicious consumers who hack their smart meters to illegally lower their electricity bills by falsifying their readings. These attacks on the AMI could have bad economic consequences and impair the performance of the power grid if these readings are used for managing the grids. Machine learning models can be used to detect false readings but this requires disclosing consumers’ CAT readings to the EU to evalaute the model. However, disclosing the consumers’ readings jeopardizes consumers’ privacy due to the fact that these readings can reveal sensitive information about consumers’ lifestyles, e.g., their presence or absence, the appliances they use, etc. The problems of detecting power theft while protecting the consumers’ privacy in CAT AMI is investigated in this paper. First, a dataset of actual readings to generate a benign dataset is developped followed by proposing new cyber-attacks tailored for CAT AMI to generate malicious samples. Then, two deep-learning detectors using a baseline model (CNN) and a CNN-GRU model are trained to detect power thefts in CAT AMI. To preserve consumers’ privacy, the paper develops an approach to enable the EU to evaluate the detector using encrypted data without being able to learn the readings. Extensive experiments were carried out to assess our proposal, and the results indicate that our proposal is capable of accurately identifying malicious consumers with acceptable overhead while preserving the privacy of the consumers. Specifically, comparing to CNN model, our CNN-GRU model increases the detection rate from 93.85% to 97.14% and HD from to 87.7% to 94.28%, respectively.