Enhancing Predictive Maintenance of Urban Streetlights through Hyperparameter-Tuned Machine Learning Techniques

In recent years, the application of machine learning techniques to urban infrastructure management has seen a significant surge, particularly in predictive maintenance of street lighting systems. This study proposes and rigorously evaluates a machine learning framework for the fault detection and classification of streetlights, leveraging Random Forest and AdaBoost algorithms, both with and without hyperparameter tuning. A comprehensive analysis was performed using a real-world dataset comprising over 34,000 observations, encapsulating electrical and environmental parameters critical to streetlight performance. The experimental results reveal that hyperparameter optimization, conducted via GridSearchCV, substantially enhances model accuracy and generalization. In particular, the Random Forest model, when subjected to systematic tuning, achieved the highest classification performance with an overall test accuracy of 84.94 percentage, markedly outperforming other models. Detailed performance metrics, including precision, recall, and F1-score across multiple fault categories, highlight the superior robustness of the optimized Random Forest approach, especially in handling class imbalance and complex fault patterns. This research not only advances the methodological landscape of predictive maintenance for urban lighting systems but also offers actionable insights for municipal planners seeking to implement intelligent, resilient, and cost-effective infrastructure management solutions.