A Spatial Analysis Methodology Based on Lazy Ensembled Adaptive Associative Classifier and GIS for Examining the Influential Factors on Traffic Fatalities

Abstract

Analyzing the influential factors of traffic accidents has been a hot topic in city management. Most existing literature in this domain implemented linear based sensitivity analysis in statistics to study the problems. However, the linear assumption limits their model performance and therefore interferes with the detection of influential factors. Recent studies started to use nonlinear machine learning methods to explore the problem. One of the most popular ways is the association rule analysis. Based on the Support and Confidence value, researchers were able to identify the top influential factors. However, (1) the identification of the thresholds for Support and Confidence has not been well solved in related studies. This study, therefore, proposes Lazy ensembled adaptive Associative Classifier to tackle this problem. Besides, (2) most of the existing literature only analyzed the general relationships between the influential factors and the traffic fatality but did not further investigate their spatial connections. Those studies could not answer specific questions like 'which region should be focused more on alcohol control?', or 'where requires more attention on motorcycle control?'. This study combines the road-based GIS analysis and the results from association rule analysis to spatially analyze the relationships between the impact factors and the traffic fatalities. Specific suggestions on city management and traffic control were proposed thereafter.