Accurate malaria detection is crucial for effective disease management, particularly in regions with limited medical resources. Deep learning models have shown promising results in automated diagnosis, yet real-world deployment often faces challenges such as computational cost and model interpretability. This study evaluates multiple deep learning architectures—VGG16, ResNet50, InceptionV3, MobileNetV2, and DenseNet121—on the publicly available National Institutes of Health (NIH) malaria cell image dataset (27,558 images), and enhances their performance using stacking ensemble learning with different meta-learners. Among individual models, DenseNet121 achieved the highest accuracy of 88.00%, while MobileNetV2 had the lowest at 84.80%. Implementing stacking with logistic regression as the meta-learner improved accuracy to 89.40%, while random forest increased it to 90.10%. The best performance was achieved with XGBoost as the meta-learner, attaining an accuracy of 91.20%, precision of 92.10%, recall of 90.80%, and an F1-score of 91.40%—representing a 3.2% accuracy improvement over the best individual model. The classification report further confirms superior performance in distinguishing infected and uninfected cases. These results highlight the potential of stacking with advanced meta-learners to support health workers in rapid, reliable malaria diagnosis, ultimately aiding timely treatment, and improving patient outcomes in clinical and field settings.
30 September 2026 in VFAST Transactions on Software Engineering
Conventional intrusion detection systems (IDSs) face increasing complexity in network traffic and unpredictable cyberattacks, particularly with fixed supervised classifiers and ambiguous observations. This study introduces DRL-ECTD (Deep Reinforcement Learning-Enhanced Cyber Threat Detector), a hybrid binary IDS that combines probability estimates from Random Forest (RF) and Artificial Neural Network (ANN) classifiers with a cost-sensitive decision-refinement mechanism using Deep Q-Network (DQN). High-confidence predictions are retained, while predictions within a preset uncertainty interval are passed to the DQN for secondary evaluation. Individual NSL-KDD records are treated as independent decision states, with DQN used for reinforcement-based refinement rather than temporal network-state modeling. On the NSL-KDD test set, DRL-ECTD achieved 78.6% accuracy, 80.4% precision, 76.3% recall, 78.3% F1-score, and 0.988 AUC. The ANN baseline achieved slightly higher accuracy (79.2%) and F1 score (79.1%). DRL-ECTD outperforms the other tested models in accuracy and F1-score and the baseline RF and standalone DQN in precision. SHAP was used for feature attribution and model interpretation. A proof-of-concept Flask service interface demonstrated deployment feasibility, although latency, throughput, computational overhead, and resource utilization were not experimentally benchmarked. The results provide initial insights into hybrid confidence-based decision refinement and motivate further experiments with multiple classes, temporal order, and deployment for intrusion detection.
1 August 2026 in IAES International Journal of Artificial Intelligence (IJ-AI)
Distributed denial of service (DDoS) attacks continues to pose serious risks to modern networks, with their growing intensity making early detection both critical and challenging. Conventional machine learning (ML) models often struggle with the nonlinear and highly dynamic nature of attack traffic, which motivates the use of advanced architectures. In this study investigate a transformer-based classifier for DDoS detection on the CIC-DDoS2019 dataset. The workflow included preprocessing, feature scaling, and domain-guided feature selection. Logistic regression (LR) was employed as a baseline, achieving 92.1% accuracy and F1-score of 0.90, thereby revealing the limitations of linear models. The transformer, after hyperparameter tuning and 5-fold cross-validation, reached an average accuracy of 99.95% with precision, recall, and F1-scores all above 99.9%. The model demonstrated stable convergence and generalization across folds. These results highlight the strength of attention mechanisms in capturing feature dependencies, while also pointing to future directions such as real-time deployment, explainability, and resilience to zero-day attacks.
1 September 2026 in International Journal of Advances in Applied Sciences
Machine learning and computer-assisted disease detection are two technological innovations that have significantly improved medical advancement, and recent research has demonstrated their effectiveness. Liver cancer is one of the important causes of cancer-related deaths internationally. Tumor detection in the liver and its classification is challenging due to poor accuracy and high processing requirements available in the existing techniques. In these, there is a loss of edge information and the quality of the images considered. To address these issues in the segmentation and classification of liver cancer, a novel feature extraction method and algorithm called the novel multilevel local region (MLR)-based sparse shape composition model (NMLR-SSC) were developed. This innovative technique identifies the existence of tumors on abdominal computed tomography (CT) images. The input images were obtained from the 3D-IRCADb-01 dataset. The algorithm showed an improvement of 0.22%. When compared to other classifiers. The proposed algorithm showed 100% specificity, sensitivity, and a 98% accuracy rate in detecting the liver tumor.