Analysis of Time-Based Public Transport Demand Prediction Using OPTUNA Framework

Abstract

Buses are the most popular and easy mode of transportation in all over the world. The state government operates bus service in all routes with low-cost fare. Traffic congestion has risen at an alarming rate due to an increase in the number of automobiles. Travel time have increased as a result, while accessibility and mobility have worsened. The primary challenge encountered by passengers is the absence of information regarding bus numbers that are accessible on a certain route and the approximate time of bus departure.  The delay in bus operations, could have several reasons which are inclement weather, traffic jam, and breakdowns. Neither the arrival time of the bus nor the delay are known to the people waiting at the bus stop. In order to address this problem, encouraging the usage of public transportation seems to be a feasible way. Over the past ten years, prediction of bus arrival time has become a fascinating subject around the world. In the transportation sector, Machine Learning (ML) technologies have already shown great promise and have additionally shown to yield a larger return on investment than traditional methods. In this research work , authors propose and develop predictive models to predict public transport demand for passenger transit based on bus arrival time. The dataset shows the proportion of buses operated by the Rochester-Genesee Regional Transportation Authority (RGRTA) that arrive on time. Initially, lazy predict classifier is used for solving regression-based dataset for predicting the bus demand in On-time based passenger transit. Based on the examination of lazy classifiers, the Decision Tree Regressor (DTR) has been identified as the best model. It is assessed using the most advanced hyperparameter optimization framework (OPTUNA). The proposed OPTUNA based DTR which is utilized to identify On-time performance of bus services-based passenger transit. Using OPTUNA for search is an efficient and beneficial approach considering the search speed and the improvement in model accuracy. According to the experimental data, the proposed approach performs better where the R-squared score is 0.9878 with best hyperparameter to be optimized.