BER-XGBoost: Pothole Detection based on Feature Extraction and Optimized XGBoost using BER Metaheuristic Algorithm

Abstract

Within the realm of intelligent transportation systems, the imperative challenge of pothole detection assumes a pivotal role in ensuring road safety and upholding infrastructure integrity. This research undertaking meticulously navigates the intricacies of automated pothole detection, employing a nuanced and multifaceted approach. The dataset, comprising over 300 meticulously labeled images of roads with and without potholes, constitutes the cornerstone of our investigation. By leveraging the robust GoogLeNet for feature extraction and orchestrating the optimization of XGBoost through the Al-Biruni Earth Radius Metaheuristic Algorithm, our proposed methodology exhibits a commendable efficacy in discerning road anomalies. The outcomes elucidate the efficacy of the implemented strategies, with BER-XGBoost emerging as a preeminent performer, achieving an accuracy rate of 96.01%. This model not only attains superior accuracy but also manifests a comprehensive array of metrics, including sensitivity, specificity, positive predictive value, negative predictive value, and F-score. Rigorous statistical analyses, encompassing ANOVA and the Wilcoxon Signed Rank Test, furnish empirical substantiation of the consequential nature of our methodologies. In conclusion, this study not only contributes practical insights to the pertinent field but also stimulates pivotal inquiries regarding the ramifications of optimization strategies and the intricate role played by feature extraction in the domain of automated pothole detection. This research propels the ceaseless evolution of intelligent systems, effectively bridging the chasm between technological progressions and real-world applications, thereby augmenting road safety and fortifying infrastructure management.