Enhancing IoT Device Security: Predicting and Analyzing Reconnaissance Attacks Using Flags and Time-Based Attributes

Nazia Sharmin, Christopher Kiekintveld

Abstract: This study delves into the comprehensive analysis of reconnaissance attacks targeting IoT devices, leveraging time-based characteristics and flag attributes exclusively. Our approach involves the strategic selection of a representative sample using Bayesian optimization to identify an optimal flow duration range. This selection not only enhances the efficiency of our analysis, as demonstrated by Kernel Density Estimation (KDE) plots, but also yields improved accuracy. By taking into account the optimal flow duration range, we present a methodology for analyzing and predicting reconnaissance attacks by constructing a sub-model and harnessing a broader multiclass model encompassing various attack types. Our methodology showcases high performance in both prediction accuracy and analytical capabilities, making significant strides in the realm of IoT device security assessment.