Driving Performance Indicator (DPI) to Classify Distracted Driving Conditions in the Elderly Females

Abstract


 
 
 The driving performance and distracted driving conditions are among the key elements in road safety assessments. Quantitative measurements were reportedly used to characterize the driving performance. However, the main Driving Performance Indicators (DPI) that evaluate and distinguish the distracted driving conditions are unknown. The present study aimed to select the top three DPIs that best classify three levels of driving distractions targeting the elderly female group based on the publicly available database. Data involved eight driving session records of 15 subjects (female of elderly age group) captured in 62,284 instances x 6 attributes (5 DPI and 1 DriveCondition class) that were extracted based on the inclusion criteria set. The DPI features were selected based on the Correlation-based Feature Selection (CFS) and CorrelationAttributeEval (CA) algorithms of WEKA 3.8, reasoned by DPI's Pearson's correlation results. The Simple Random Undersampling approach was used to resolve the class imbalance state. The 'All' and 'with DPI feature selection DPI' (CFS and CA) datasets were classified using 1NN and J-48 algorithms at 10-fold cross-validation mode into three predefined classes of distracted driving conditions (Relax, Moderate and Intense). Classification accuracies achieved from 'All' and 'with DPI feature selection DPI' (CFS and CA) datasets were 66.10% to 68.86% (1NN) and 68.50% to 71.01% (J-48), respectively. The main DPI subsets nominated by CFS: {Speed, Acceleration, Steering, Laneoffset} and CA: {Speed, LaneOffset, Acceleration} each decreased classification accuracy from All datasets by a minimal 0.4% to 2.8% each. Findings demonstrated that Speed, Acceleration, and Lane Offset were high-ranked DPIs that sufficiently distinguished driving distraction classes for the elderly female drivers.