Machine Learning Based Fault Determination in Near-crash/Crash Scenarios

Project Description The objective of this project was to develop a fault determination classifier which determines whether the subject vehicle or the surrounding vehicle is responsible for the near crash/crash event. Data Request Scope Data for all crashes and near-crashes found on InSight were requested. Data Specification Time-Series Speed (GPS and Network) Acceleration/Deceleration (x and y axis) Odometer Pitch Rate Roll Rate Steering Wheel Status Brake Pedal Status PRNDL (Gear) Details Turn Signal Cruise Control Lane Markings, Lane Width Alcohol Flag Temperature, Interior Details Illuminance, Ambient Details ABS Activation Details Event Detail Table Fault Crash Severity (1 and 2) Precipitating Event Hands on the Wheel Vehicle Contributing Factors Details Secondary Task Lighting Number of Objects Number of Motorists Incident Type Trip Summary Table Cell Phone Flag Time Wipers Used Lights Usage Percentage Number of Lateral Accels > Threshold Distance with Lead Vehicle Time with Lead Vehicle Time Where Headway 1.0 - 1.5 s Time Where Headway 3.0 - 3.5 s Trip Start UTC Hour of Day Vehicle Details Model Year