Exploring the Potential of Feature Selection Methods for Effective and Efficient IoT Malware Detection

The growth of technology and the internet has brought significant negative impacts, especially in terms of vulnerability to cybercrime, such as malware, in the Internet of Things (IoT) environment. IoT malware is becoming an increasingly serious threat as internet-connected devices become more widespread and diverse, affecting everything from home smart devices to motor vehicles. Previous studies have introduced many malware detection models. Still, none has focused on analyzing the relationship between data features. This study aims to explore the relevance between features and improve the effectiveness of machine learning in detecting and classifying IoT malware. The proposed method employs feature selection techniques, including Chi-Square, Pearson Correlation, ANOVA, and Graph Feature Selection. The experiment shows the combination of Pearson Correlation and Gradient Boosting obtained the best result with an Accuracy of 99.98%. Thus, the result obtained optimal Precision, Recall, and F1-Score of 99.97%. Random Forest demonstrates strong performance without feature selection, achieving an Accuracy of 99.91%, Precision, Recall, and F1-Score of 99.90%.