Deteksi Serangan Siber Jaringan Internet of Things Menggunakan Random Forest pada Dataset IoT-23

Abstract

Internet of Things networks are vulnerable to malicious traffic because connected devices often operate with limited resources and weak security configuration. This study proposes a binary cyberattack detection model for Internet of Things traffic using Random Forest and a labeled IoT-23 connection log subset. The research used a quantitative experimental method with the uploaded conn4_log_labeled.csv file. After metadata cleaning, the dataset contained 156,103 valid records, consisting of 4,536 benign records and 151,567 malicious records. Identifier attributes, detailed labels, timestamps, and raw Internet Protocol addresses were removed to reduce data leakage. Numerical features were converted and imputed using the median, while categorical features were encoded through one-hot encoding. The dataset was split into 80 percent training data and 20 percent testing data using stratified sampling. Random Forest used 100 trees, Gini criterion, balanced class weighting, and random_state 42. The model achieved 99.9968 percent accuracy, 99.9968 percent weighted precision, 99.9968 percent weighted recall, 99.9968 percent weighted F1-score, and 99.9999 percent ROC-AUC. The confusion matrix showed 907 true benign records, 30,313 true malicious records, zero false positives, and one false negative. The most influential features were history and conn_state. These findings show that Random Forest provides a strong and interpretable baseline for binary Internet of Things intrusion detection.