DriveCheck: A Driving Behavior Monitor

This paper introduces the DriveCheck driving behavior monitor, a web application for the evaluation of risky driving behaviors. The tool uses a random forest supervised learning algorithm to classify normal and risky events based on acceleration, braking, and turning information. Initially, a public Kaggle dataset labeled by driving level was used. Subsequently, a proprietary dataset was collected from 27 real urban trips in Lima, where selected maneuvers (sharp turns, sudden acceleration, and harsh braking) were induced under controlled conditions to calibrate the model. The results showed a risk classification accuracy of 97%, validating the robustness of the model and its applicability in real urban settings (field tests in Lima).