ROBUST AND LIGHTWEIGHT CONTRASTIVE RIDGE FEATURE SELECTION FOR EFFICIENT IOT MALWARE DETECTION

Abstract

The connected smart devices, sensors, and systems in the Internet of Things (IoT) enables flawless data exchange and supports automation in healthcare, industry, and smart cities. Though the life gets convenient with increasingly connected world , it also introduces security risks, hence strong protection is essential. The resource constrained IoT devices demands a lightweight malware detection ensuring balance between detection efficacy and computational cost. The selection of appropriate features is required in this regard. A novel method Contrastive Ridge Feature Selection (CRFS) is proposed in this paper , which make use of Ridge regression coefficients to identify the most discriminative features distinguishing between malicious and benign data. The feature ranking of CFRS is according to their contrastive importance between benign and attack classes. We conducted an extensive evaluation of CRFS on IoT malware datasets such as N-BaIoT, Bot-IoT, KDD Cup 2018, and Edge-IIoTset, which demonstrates its effectiveness.The CRFS-based feature selection not only boosted the accuracy of the Linear SVC model—achieving 97.15% on Bot-IoT, 94.85% on Edge-IIoTset, 93.69% on KDD Cup, and 85.06% on N-BaIoT—but also significantly reduced training time by up to 85.84% and minimized feature redundancy to as low as 0.71%, making it highly suitable for real-time IoT malware detection.