Using Machine Learning to Identify Clustering Patterns of Traffic Accidents

Between 20 and 50 million people suffer non-fatal injuries from traffic accidents each year, while more than one million of these result in death. Road traffic accidents pose a significant threat not only to the economy but also to public health. The figures obtained are objective statistics and not based on personal opinions. This study aims to develop a clustering model using a machine learning approach based on the characteristics of the occurrence and number of victims of traffic accidents in Palembang City, South Sumatra Province, Indonesia. This research will analyze patterns that observe different relationships among crashes by using various clustering techniques such as K-means clustering, Gaussian mixture model (GMM), density-based (DBSCAN), hierarchical clustering, spectral clustering, and OPTICS (sorted points to identify clustering structure). Through the machine learning approach, this research attempts to bridge the knowledge gap between the factors involved in traffic accidents and the development of more effective prevention strategies. In addition, this research also provides further insight into the potential use of machine learning algorithms in analyzing and processing traffic accident data. The results show that an algorithm called spectral clustering outperforms the other algorithms with a low Davies-Bouldin score (0.3221), a high Calinski-Harabasz score (14789.9374), and a silhouette coefficient (0.7695). Spectral clustering was the best algorithm out of the six algorithms evaluated for this paper. By favoring spectral clustering as the best algorithm, this research provides a new outlook on the application of technology in the field of highway safety. The results also show that there are specific patterns of traffic accidents in Palembang City that can be identified through data analysis using clustering techniques. The implications of this research provide an important contribution to the development of strategies for traffic accident prevention and road safety improvement in the region. It is expected that the findings can assist the government and related agencies in taking more effective measures to reduce the number of traffic accidents and protect the public from the associated risks.