Comparative Analysis Of Machine Learning Algorithms On Weighbridge Data For Overloaded Truck Prediction (A Case Of Gilgil Weighbridge Station)

Abstract

Rapid increase of motorization, increased business and lack of alternative means for cargo transport leads to congestion at weighbridges resulting in delays in clearance at the axle-load control facility. Various weighbridge layouts have been recommended depending on the volume of trucks to be checked but this also have had challenges such as weighbridge breakdowns and unprecedented increase in traffic volumes. Machine learning algorithms that predict overloaded trucks with use of previous weighbridge data can contribute to the weighbridge traffic efficiency by identifying overloaded trucks that are then subjected to mandatory checks, clearing those predicted as within permissible limits. Prediction is a technique that generalizes the trends in the data that can then be applied to new instances. Data continually collected at the weighbridge facility was analyzed and wheel configuration, unit axle load and gross vehicle weight considered for prediction of overloaded trucks. The data was preprocessed and split in 2/3 and 1/3 partitions, and further in the ratio 10:1 of training and testing datasets respectively and loaded onto the Waikato Environment for Knowledge Analysis (WEKA). In this research performances of Random Forest, J 48 algorithm, Naïve Bayes, Multilayer Perceptron and PART algorithms were analyzed and a model of PART algorithm that had an accuracy of 73.1% deployed. 50 new unknown data instances were used to evaluate the model and a prediction accuracy of 88% was recorded.