Revealing crash hotspots concerning google traffic maps historical data by supervised and ensemble machine learning techniques

• Advanced machine learning models (XGBoost and SVM) predicted crash hotspots with 88 % accuracy. • Diverse data sources, including Google traffic maps and meteorological data, identified key crash factors like dynamicity. • The framework offers actionable insights for proactive crash prevention and improved traffic safety. Road crashes are common occurrences that can have serious consequences. They are influenced by a range of factors, including environmental conditions, human behavior, and road design. To prevent crashes, it is essential to identify crash hotspots. Online traffic maps are a new source of data that can provide insight into traffic patterns and help predict crashes. This study focuses on developing an accurate and practical real-time crash prediction model by leveraging machine learning techniques. Unlike previous studies that either emphasize model accuracy or real-time prediction separately, this research aims to balance both aspects, offering a feasible and efficient solution for crash hotspot identification. The key innovation of this study lies in integrating real-time dynamicity analysis from Google traffic maps with historical crash records and meteorological data to enhance both predictive performance and real-time applicability. The model, focusing on the hourly level, strikes a balance between computational efficiency and practical deployment in large-scale urban networks. Machine learning techniques, including Extreme Gradient Boosting (XGBoost) and one-class Support Vector Machines (SVM), were employed, and the results indicated that XGBoost outperformed the one-class SVM. The overall accuracy of XGBoost and one-class SVM was 88 % and 75 %, respectively, when compared to the actual crash occurrences.