Comparison of Random Forest and Decision Tree methods: Study of cyclist accident analysis

Machine learning algorithms have become increasingly popular in accident analysis, replacing traditional regression. Machine learning algorithms have a learning process and can build classification and prediction models, and can work with both large and small datasets. This study analyzed cyclist accidents that occurred at roundabouts. Two machine learning methods, Decision Tree and Random Forest algorithms, were used to classify cyclist injury severity in the dataset based on independent variables. Literature studies indicate that Decision Tree and Random Forest algorithms provide high accuracy in classification. This study aimed to demonstrate the impact of each independent variable on the injury severity by leveraging the robustness of Decision Tree and Random Forest algorithms. The results also compared the two machine learning methods. Roundabouts, which significantly reduce the injury severity for motor vehicles, do not have the same effect on cyclists. Because cyclists are disadvantaged at roundabouts, cyclist accidents were analyzed. 439 samples were used to perform the analysis. The dataset contains information on cyclist-motorist accidents occurring at various roundabouts in Newcastle, England. The data is divided into two categories: mild and severe. In the analysis, the degree of injury is the dependent variable; the independent variables are the roundabout geometric elements, sociodemographic information of the drivers, driver behavior, speed limit, and weather conditions. The dependent variable in the dataset, degree of injury, is categorical, and classification is performed. R software was used to perform the analyses. Decision Tree and Random Forest are both supervised learning algorithms. Data are split into 70% and 30% for training and testing in Decision Tree and Random Forest algorithms. A confusion matrix is used to measure classification accuracy. The confusion matrix numerically displays the values correctly and incorrectly predicted by the method used. A high algorithmic accuracy is crucial for selecting a good classification algorithm. The analysis revealed a classification accuracy of 86.15% for Decision Tree and 83.33% for Random Forest. In the literature, it is generally observed that Random Forest yields higher accuracy than Decision Tree. The lower results obtained with Random Forest in this study are thought to be due to the dataset, and the variation in the selected 70% training data each time may affect the accuracy rate. It is not appropriate to recommend a single machine learning algorithm for accident analysis. Future accident analysis studies should also test different machine learning algorithms on different datasets.