ANALISIS EFEKTIVITAS SISTEM DETEKSI INTRUSI BERBASIS MACHINE LEARNING UNTUK KEAMANAN JARINGAN LOKAL

Abstract

Rapid growth in local area network (LAN) infrastructure across educational institutions and offices introduces increasingly complex cyber threats. Conventional rule-based Intrusion Detection Systems (IDS) often fail to identify zero-day attacks and exhibit high false alarm rates. This study evaluates the effectiveness of an artificial intelligence-based IDS by comparing three machine learning algorithms: Random Forest (RF), Support Vector Machine (SVM), and Naive Bayes (NB) using the UNSW-NB15 benchmark dataset. The pre-processing pipeline includes data cleaning, Min-Max normalization, categorical encoding, and feature selection via Recursive Feature Elimination (RFE), reducing the feature space from 42 to 15 dominant attributes. Performance was evaluated using Accuracy, Precision, Recall, F1-Score, and Inference Latency per packet. Experimental results demonstrate that Random Forest significantly outperforms other models, achieving 98.2% Accuracy, 97.9% Precision, 98.0% Recall, and a 97.9% F1-Score, with a low inference latency of 0.03 seconds per packet. In comparison, SVM achieved 95.1% Accuracy (0.12s latency), while Naive Bayes reached 89.4% Accuracy with the fastest latency of 0.01s. This research confirms that combining Random Forest with RFE provides an effective and efficient foundation for adaptive IDS deployment in resource-constrained LAN environments.