TRAFFIC ACCIDENT PREDICTION USING MACHINE LEARNING

This report presents the results from the research study on applying large scale data mining methods into analysis of traffic accidents on the finnish roads. The data sets collected from traffic fatal accidents are huge, multidimensional, and heterogeneous. Moreover, they may contain incomplete and erroneous values, which make its exploration and understanding a very demanding task. The target data of this study was collected by the finnish road administration datasets. The intention is to investigate the usability of robust clustering, association and frequent itemsets, and visualization methods to the road traffic accident analysis. While the results show that the selected data mining methods are able to produce understandable patterns from the data, finding more fertilized information could be enhanced with more detailed and comprehensive data sets. Machine learning algorithm takes accident frequency count as a parameter to cluster the locations. Then we used association rule mining to characterize these surface condition. The rules revealed different factors associated with road accidents at different drunk and drive with varying accident frequencies. The association rules for high-frequency accident location disclosed that intersections on highways are more dangerous for every type of fatal accidents.