An Interactive Web Tool for Vehicle Telematics Visualization and Supervised Anomaly Detection

Modern commercial vehicles generate massive streams of onboard sensor data, yet fleet operators and data scientists lack accessible tools that couple interactive visualization with machine learning for behaviour analysis. This paper presents an interactive, browser-based tool that integrates time-series visualization of vehicle telematics with a supervised-learning workflow for anomaly detection. Built with the Python Dash framework and scikit-learn, the tool organizes the analysis into four linked stages: (i) data visualization, where users filter sensor signals (e.g., engine speed, fuel temperature, GPS trace) by date and range and inspect them on synchronized scatter plots and a map; (ii) manual labelling, where users select data segments and annotate them as normal or abnormal; (iii) modelling and evaluation, where classifiers (K-Nearest Neighbours, Decision Tree, Neural Network, Support Vector Machine) are trained on the labelled segments and compared by Accuracy, F1-score, and average precision; and (iv) automatic labelling, where the best model is applied to unlabelled data to flag anomalies at scale. We demonstrate the tool through two case studies on real vehicle sensor logs: relating abnormal engine-speed spikes to road intersections, and detecting fuel-temperature outliers. In the demonstration, K-Nearest Neighbours achieved the best classification scores across all three metrics. The tool lowers the barrier to exploratory telematics analysis by unifying visualization, labelling, and modelling in a single shareable web application.