Logistic Multilayer Perceptron Neural Network Classification based Driving Behavior Prediction through Vehicle Motion Data

Traffic accidents are caused in effect through driver's negligent driving habits. Driving aggressively is a significant component that can rapidly result in risky driving. Safe and efficient erratic driving detection techniques, which allow for the creation of driving behavior analysis and early warning systems, are critically required in order to decrease traffic injuries and enhance road safety. With this motivation, this work uses “Vehicle motion dataset” which contains the motion information gathered by a phone affixed to the vehicle's accelerometer and gyroscope. This paper aims to forecast the driving behavior of the driver by suggesting “Logistic Multilayer Perceptron Neural Network (LMPNN)” uses the vehicle motion dataset retrieved from the KAGGLE database. The vehicle motion dataset consists of 8 features that collectively contain 3084 driving movement data. The LMPNN begins by interpreting exploratory data, which had been processed with insufficient data. The information was then divided in an 80:20 ratio for testing and learning. Geforce Nvidia V100 Graphics card processor with 30 learning iterations, each subject to a 64-batch threshold, was used to code Python scripts. To test the effectiveness of predicting driving behavior, the dataset was run through all the classifiers and a multilayer perceptron with relu, identity, tanh, and logistic activation layer. Execution results portrays that suggested LMPNN that is MLP model with logistic layer having 99% of Accuracy when compared to other existing classifiers.