Imbalanced Learning Analysis for Driving Behavior Prediction Using Naturalistic Driving Data

• DOI: 10.2139/ssrn.4532416
• Publication Date: 2023-08-09T15:00:18Z
• Authors: Antonis Kostopoulos; Thodoris Garefalakis; Eva Michelaraki; Christos Katrakazas; George Yannis
• Keywords: Multilayer perceptron; Gradient boosting; F1 score; Boosting

Human behavior has a major role in severe road injuries, making safety enhancement crucial. Recent studies have focused on driving analysis through the development of machine and deep learning models to detect quantify relations between different features. This research aims contribute field's background knowledge by employing various techniques detect, classify, predict dangerous harsh events using algorithms designed for imbalanced datasets. study evaluates influence specific features occurrence severity events, utilizing feature selection. For this purpose, five classification were developed classify behavior, namely Random Forest (RF), Gradient Boosting (GB), Extreme (XGBoost), Multilayer Perceptron (MLP), K-nearest Neighbors (kNN), are employed. GB MLP outperform others, achieving recall rates approximately 67% 68% both acceleration braking events. The input variables process include total distance, duration, average speed, speeding score, mobile use score. Furthermore, K-means clustering method determine levels, emerged that 48.82 accelerations per 100 km 45.40 brakings optimal threshold safe risky behavior. offers promising alternative approach predicting road, examine human providing valuable insights effective measures.