Efficient Intrusion Detection with DL-Protect+: Balancing Accuracy, Interpretability, and Resource Efficiency

Efficient and scalable intrusion detection systems (IDS) are crucial for keeping up the ever-changing world of enterprise Cybersecurity besides promptly mitigating threats.Finding hostile actions and any security holes in a system or network is what cyber threat detection is all about.It monitors, detects, and alerts on anomalies using advanced technologies such as machine learning, behaviour analysis, and signature-based methodologies.In avoid data loss, service disruption, besides unauthorised access, early detection is crucial.In today's digital infrastructures, it is an essential part of Cybersecurity plans.This study presents DL-Protect+, a compact deep learning system developed for large-scale network threat detection with good performance.The model incorporates explain ability mechanisms using Grad-CAM, attention heat maps, and a hybrid optimisation algorithm that combines Falcon Search and Willow Catkin for feature selection; a Temporal Fusion Layer (TFL) coupled with Depth wise Separable Convolutional Neural Networks (CNNs) for classification; and various other methods.DL-Protect+ surpasses more conventional models such as CNN (accuracy: 87%), LSTM (88%), BiLSTM (89%), GRU (88%), besides Transformer (90%), with impressive results of 95.4% accuracy, 0.96 F1-score, and 0.98 AUC-ROC.It is ideal for deployments at network's periphery because of its optimum computational efficiency, which is demonstrated by its 8.7 MB model size, 1.4 ms inference time/sample, and 0.35 J/sample energy consumption.With a small quantisation decrease of 1.1% and a fast cold start latency of 220 ms, deployment measurements show the lowest ONNX conversion time of 1.2s.In addition, DL-Protect+ is resilient with an accuracy of 92.1% across datasets and effective performance even when faced with noise and domain adaptability.Ablation experiments confirm the value of each module, demonstrating that the detection performance is substantially improved by the TFL and feature selection components.An innovative, interpretable, and resource-efficient intrusion detection system (IDS) that can fulfil real-time needs across enterprisegrade infrastructures is introduced in this study as DL-Protect+.