Road Condition Detection Techniques: A Comparative Analysis based on Supervised Machine Learning Techniques

With the advancement in autonomous vehicle and use of intelligent transportation system, road accidents are also increasing at higher rate. Road safety becomes a one of the major concerns for the transportation system and government. Road accident and road death are gaining highest percentage ration in number of deaths. Potholes, road bumps and slippery roads are the prime reason for the road accident. To take the prevention against road accident many technology-based solutions were provided by the research community. These solutions are based on machine learning techniques and smart phone-based technologies. In this paper, machine learning techniques were used to predict the road conditions. Kaggle data set of the road condition is considered for the experiment. Accuracy of Random Forest, Decision Tree, K Neighbors, Naïve Bayesian, Suport Vector Machine and Multi-layer Perceptron Classifier is compared using the vibration dataset of road network. Maximum accuracy of 91.8 percent and F1 score of 97 percent was achieved by the random forest classifier. Whereas minimum accuracy value 25 percent was achieved by the Complement Naïve Bayesian Classifier. Accuracy, F1 score, Recall, Precision and other calculated results can be used by the intelligent transport system to generate the alert to driver about the road condition and further these algorithms can be used by the road authority to identify the road conditions and then to take corrective action like road repair and maintenance.