This paper describes clustering and investigation of human driving behavior with shoe type in urban roads using autoencoder. The analysis of driving patterns for different shoe types is important as it has been known to affect safe driving due to the braking distance change. To this end, we analyzed the effect of the type of shoes on vehicle driving by using an autoencoder to train the data acquired from actual vehicle driving with two types of shoes. With successfully trained data, the driving characteristics have been clustered on various urban driving scenarios when a preceding vehicle exists. The validity and impact of the clustering results on safe driving were verified through collision risk analysis in order to investigate the safety effects of shoe types. It has been shown from vehicle tests that the proposed clustering analysis with probabilistic risk assessment presents a clear correlation between footwear choice and driving safety, with the maximum collision probability was reduced by 23%, and the maximum collision time was improved by 0.4 seconds when driving shoes were worn.