Analysis of Road Damages for Micro Mobility Vehicles Via Synthetic Data: Three-Axis Accelerometer-Based Machine Learning

İngilizce (OpenAlex)

The effect of road damages on the road surface on driver safety and comfort depends on the damping mechanism of the vehicle. Since micro mobility vehicles have small wheels, road damage affects them with varying severity. This study aims to determine road damage based on the response of bicycles, e-bikes, and e-scooters to the road surface. In order to achieve this goal, firstly synthetic data approach is adopted. There are 10 000 samples in this data set and it was produced on Google Colab based on Python. These samples simulate data collected with a three-axis accelerometer. In order for the road damage distributions to represent the real world, flat roads (undamaged), cracks and potholes are determined as 7 000, 2 000 and 1 000 samples, respectively. In order to prevent the distribution from being biased and to eliminate the overfitting problem, unbalanced class distribution and sensor noise are simulated. Random Forest algorithm is used for the classification of damages. The classification accuracy rate of the damages is 95%. In addition, the K-Means clustering algorithm helps analyze how each micro mobility vehicle type responds to road damages. The Silhouette Score is 0.543, which shows how intertwined the clusters are and how separate they are from each other. The results confirm that the proposed approach integrates well with real-world data. To validate model performance, researchers should collect real accelerometer data alongside simulated data.