An Adaptive Real-Time Threat Severity Assessment System Using Machine Learning for Low-Latency Decision Support in Defense Environments

This paper presents an adaptive Threat Severity Assessment System (TSAS) designed to support real-time decision-making in defense environments. Existing systems primarily focus on detection and tracking, often lacking a dedicated layer for severity evaluation and prioritization. This limitation can lead to inefficient response strategies in time-critical scenarios.The proposed TSAS framework operates as an intermediate intelligence layer that processes heterogeneous sensor inputs and classifies threats into multiple severity levels using a feed-forward neural network. The system integrates entropy-based concept drift detection and incremental learning to maintain performance under evolving threat conditions. A composite severity score is generated by combining model predictions with proximity-based heuristics.The framework is evaluated using a high-fidelity synthetic dataset, achieving an overall classification accuracy of 96.4% and a millisecond-level end-to-end latency of approximately 4.5 ms under controlled hardware conditions. Experimental results demonstrate improved prioritization capability and robustness compared to traditional rule-based, fuzzy logic, and static machine learning approaches.These findings suggest that TSAS provides a computationally efficient and adaptable solution for real-time threat severity assessment, with potential applicability in defense systems, drone monitoring, and critical infrastructure protection.