MACHINE LEARNING-BASED INTRUSION DETECTION FOR RESOURCE CONSTRAINED NETWORKS

ABSTRACT The swift expansion of networked devices in homes and small offices has increased cybersecurity threats, yet conventional intrusion detection systems (IDS) frequently require high processing power, rendering them inappropriate for implementation in situations with limited resources. In this paper, a lightweight machine learning-based intrusion detection system (IDS) tailored for small networks with constrained memory and processing power is designed, implemented, and evaluated. The study employed a comprehensive methodology utilizing the CICIDS2017 dataset which underwent extensive preprocessing, including feature scaling, data cleaning, normalization, encoding, and Principal Component Analysis (PCA) to reduce dimensionality. Decision Tree, Random Forest (pruned), Naive Bayes, K- Nearest Neighbors, and Ridge Classifier machine learning algorithms were used and evaluated with recall, accuracy, precision, and F1-score as performance indicators. The experimental results showed that the Random Forest, achieved an outstanding performance with 99.44% accuracy, 99.38% Precision, 98.11% Recall, an F1-score of 98.6%, and struck the greatest compromise between accuracy and efficiency. The interactive dashboard helps in real time monitoring and visualization. The IDS offers a resource-efficient, scalable, and user-friendly security solution designed for small businesses and personal network environments by combining pruned machine learning techniques with useful deployment frameworks. Future research should extend functionality to the cloud-based environment and deep learning approaches. Keywords: Intrusion Detection System, Machine Learning, Small Networks, Predictive analysis, lightweight models, anomaly detection. Word Count: 216