Driving Behavior Classification Using Deep Learning In Case Of National Transport

Abstract

The Classification Of Driver Behavior Is Critical Area Of Research With Significant Implication Of Road Safety, Autonomous Driving And Intelligent Transportation System. There Have Been Researches That Related With Driver Behavior Classification That Focus On Real Time Driver Drowsiness Detection Using CNN-LSTM Model, Drowsiness Classification System Using CNN With EEG Signals And Abnormal Behavior Detection Using Various Machine Learning Methods Like Random Forest,KNN,Na??Ve Bayes And SVM.Which Lacks Model Reliance Only With Facial Features Which Lead Limitation In Its Robustness, Which Require Manual Data Validation, Making The Process Time Consuming And Limitation With Small Sample Size Respectively. This Thesis Explores The Application Of Deep Learning Techniques To Classify Driver Behavior Using Dataset From Transport Company That Is Extracted From GPS Device Configured To The Trucks Includes Various Violation Records Over Six Month Period. The Dataset Comprises Time-Stamped Records Of Violation Categorized Into Different Severity Levels, Labeled As Less-Severe, Severe, And Fatal. Our Approach Involves Rearranging And Pre-Processing The Data To Handle Missing Values, Standardizing Features, And Encoding Categorical Variables. We Constructed Custom CNN, LSTM, And GRU Models To Capture Complex Patterns From Driver Behavior Dataset. Besides, We Compared The Performance Of The Deep Learning Models With Selected Conventional Machine Learning Methods. From The Experimental Results, It Is Clear That Deep Learning Method LSTM Have Shown Good Performance Compared To Other Methods. Future Work Could Extend This Research By Incorporating Environmental Factors Such As Traffic And Weather Conditions During Violation To Further Enhance The Ability Of The Models Effectiveness On Classification Capability.