Analyzing the Effectiveness of Imbalanced Data Handling Techniques in Predicting Driver Phone Use

Distracted driving leads to a significant number of road crashes worldwide. Smartphone use is one the most common causes cognitive distraction among drivers. Available data on drivers’ phone presents an invaluable opportunity identify main factors behind this behavior. Machine learning (ML) techniques are effective for purpose. However, potential and usefulness these limited, due imbalance available data. The majority class instances collected drivers who do not their phones, while minority those phones. This paper evaluates two approaches handling imbalanced datasets driver use. These methods include oversampling undersampling. effectiveness each method was evaluated using six ML techniques: Multilayer Perceptron (MLP), Support Vector (SVM), Naive Bayes (NB), Bayesian Network (BayesNet), J48, ID3. proposed were also three Deep Learning (DL) models: Arch1 (5 hidden layers), Arch2 (10 Arch3 (15 layers). used in document through direct observation study explore set human, vehicle, surface characteristics. results showed that all methods, as well DL achieved balanced accuracy values both classes. ID3, MLP outperformed rest scenarios, with ID3 achieving slightly better accuracy. provided good performances, especially undersampling classification performed best undersampled It concluded has highest impact cell use, followed by age group, gender, vehicle type, and, finally, seatbelt usage.