Design and development to flayered frame work for storage and retrieval of multi dimensional traffic data

Abstract

The exponential rise in urban populations has led to increased private vehicle ownership, causing widespread issues like traffic congestion, delays, stress, and heightened pollution levels, incurring significant expenses for households and economies. Forecasts predict a further increase in costs, underscoring the necessity for smarter big data technologies to manage traffic efficiently and alleviate congestion. Intelligent Transportation Systems (ITS) offer a solution by integrating sensors, communication tech, and advanced applications. However, managing the vast and varied traffic Big Data presents challenges requiring innovative processing and integration methods. The thesis addresses the persistent challenge of missing traffic loop detector data, enhancing network-wide data quality through a statistically rigorous imputation methodology. This method, utilizing Dual-stage Error-corrected Boosting predictor (DEB) and Multivariate Error Prediction based Uncertainty Estimator (MEPUE), outperforms existing methods, ensuring true statistical properties with lower computational complexity and generating accurate prediction intervals. Additionally, it introduces a multi-layered framework for managing multi dimensional traffic data, designed to handle, process, and derive insights from extensive transportation-related data within ITS. Emphasizingreal-timedataacquisitionandhistoricaldataanalysis,this framework aims to mitigate uncertainties in integrating various Big Data technologies, enhancing security protocols, seamless data flow, and aiding in traffic management, incident prediction, and understanding commuter behaviour. Lastly, the thesis focuses on traffic speed forecasting within ITS, presenting a novel approach utilizing gradient boosting trees to model complex spatio-temporal correlations. This method enhances accuracy and computational efficiency, demonstrating superior adaptability and prediction stability over longer horizons, considering autocorrelations to improve statistical distributions. newline