DEVELOPMENT OF TRAFFIC VOLUME FORECASTING MODEL USING MULTIPLE REGRESSION AND PRINICPAL COMPONENT ANALYSIS

Abstract

This paper is focused in development model for traffic volume forecasting in Anamorava region. Demographic and socioeconomic macro variables (independent variables) are identified at country and region level which have an impact on generation of traffic volume (dependent variable), using dataset for the period 2004-2016. Multiple Regression Analysis (MLR) method was used to build dependency between variables and model development. In order to increase forecasting capabilities of the MLR model, it was necessary to eliminate high correlation between variables (multicollinearity phenomenon). A new methodology was used, with included Principal Component Analysis (PCA) by transforming original variables in Principal Components (PCs). With the combination PCA and MLR methods are developed hybrid model as called Principal Component Regression (PCR). By employing performance indicators, it was found that the PCR model performs much better than MLR model.