Predictions of Undesirable Behaviors While Driving Part 2

The driving behavior of the driver is a major factor that can determine to what extent road accidents may occur. One of the behaviors that contributes to accidents is distracted driving and fatigue while driving, which divert attention away from driving. This has prompted many researchers to attempt to find methods of detecting these behaviors in order to reduce road accidents stemming from these causes. One of the detection methods that many researchers are interested in is using pressure sensors installed on seats and headrests. This detection method is convenient to implement as it does not require installation of any sensors on the driver. This paper is a continuation researcher study that reducing the usage of pressure sensors from the previous study [1], which reduce utilization of 9 sensor positions, down to 6 while maintaining high detection and classification accuracies. Analysis methods, including statistical data analysis, SelectKBest, and Recursive Feature Elimination (RFE) with 3 Machine Learning (ML) models. They then assessed which sensor positions had the highest importance scores, selecting the top 6 positions for further evaluation with Machine Learning to determine how much their performance differed from the original. The analysis revealed that when using only the selected 6 positions, referred to as 6 features, for training and testing with the Decision Tree model, the performance reduced by only 0.0197 percent.