Preliminary Sensing of Wrong-Lane Accidents by Comparing Random Forest with Logistic Regression, Decision Tree and SVM Algorithm for Better Accuracy

Abstract

Aim: The proposed study aims to perform detection of wrong-lane accidents using Novel Penalty Based Logistic Regression
(LR) algorithm and compare accuracy with Decision Tree (DT), Support Vector Machine (SVM) and Random Forest (RF)
algorithm. Materials and Methods: Novel Penalty Based Logistic Regression is applied on a road accident dataset that
consists of 1834 records. A Machine learning technique for the detection of wrong-lane accidents which compares Novel
Penalty Based Logistic Regression with Decision Tree, SVM and Random Forest has been proposed and developed. Sample
size was calculated as 21 in each group using G power. Sample size was calculated using clinical analysis, with alpha and beta
values of 0, 05 and 0.5, 95% confidence, 80% pre-test power and enrolment ratio is 1. The accuracy of the detection of wronglane accidents was evaluated and recorded. Results: The accuracy was maximum in detection of wrong-lane accidents using
Novel Penalty Based Logistic Regression (90.0%) with minimum mean error when compared with Decision Tree (89.88%),
SVM (89.90%), Random Forest (89.77%) and attained significance value of p = 0.02. Conclusion: The study proves that the
Novel Penalty Based Logistic Regression Algorithm exhibits better accuracy than the Decision Tree, SVM and Random Forest
in detection of wrong-lane accidents.