Temporal Evaluation of Hybrid Artificial Intelligence and Machine Learning Models for Malware Detection using The EMBER Dataset

Malware detection remains a critical challenge in computer science and cybersecurity due to the continuous evolution of malicious software, concept drift, and the declining generalization capability of machine learning models over time. This study investigates the temporal performance of hybrid artificial intelligence (AI) and machine learning (ML) models for static malware detection using the EMBER dataset. A quantitative experimental design is employed to compare Logistic Regression, Random Forest, LightGBM, and Multilayer Perceptron (MLP) classifiers with a hybrid ensemble approach. The methodology incorporates feature preprocessing, chronological data partitioning, supervised model training, and temporal performance evaluation using accuracy, precision, recall, F1-score, and the area under the receiver operating characteristic curve (ROC-AUC). Comparative analysis examined differences in classification performance and temporal degradation across the evaluated models. Confusion-matrix analysis and false-negative rates were also considered to assess classification reliability and identify potential detection errors. The hybrid framework combined tree-based learning and neural network predictions to investigate classification performance and detection consistency across temporally distinct malware samples. The significance of this research lies in addressing the limitations of conventional random-split evaluation and providing a more realistic framework for assessing the temporal generalization of intelligent malware detection systems. The findings are intended to inform the development of adaptive, scalable, and robust cybersecurity solutions capable of responding to evolving malware threats.