Classification of Mobile Usage Car Driving Activities Using Convolutional Neural Network

A traffic accident is an incident on the road that is unexpected and unintentional involving a vehicle with or without other road users that results in human casualties and loss of property value loss. The lack of awareness of the driver is a critical behavior in driving safety to deal with all possibilities that can occur while driving to maintain the safety of the driver and passengers. Several factors cause factors causing traffic accidents in Indonesia. The human factor is the most significant, with one example being using cell phones while driving. Previously, research on the introduction of human activities was conducted using several methods and approaches, but there are still several issues, such as a high false detection rate, inefficient architecture, and unsuitable data and feature use. As a result, the Convolutional Neural Network (CNN) algorithm was used in this study to classify human activities in images of car drivers using cell phones. This research collected image data from the Kaggle site, preprocessed data, classification using the CNN algorithm, and evaluation and analysis. The model has the highest accuracy of 99.4% in 5 classes of test data, safe driving, texting-right, talking on the phone-right, texting-left, and talking on the phone-left, with hyperparameter configuration, the number of batch size 16; learning rate 0.001; epochs 50; hyperbolic tangent activation function; the number of hidden neurons 32; and Adam optimizer, based on the test results and hyperparameter tuning using the Halving Grid Search. The training process requires 8.18 hours, with the best model training time at 137.66 seconds.