Evaluation for Long Term PM10 Concentration Forecasting using Multi Linear Regression (MLR) and Principal Component Regression (PCR) Models

Abstract

Air pollution in Peninsular Malaysia is dominated by particulate matter which is demonstrated by having the highest Air Pollution Index (API) value compared to the other pollutants at most part of the country. Particulate Matter (PM10) forecasting models development is crucial because it allows the authority and citizens of a community to take necessary actions to limit their exposure to harmful levels of particulates pollution and implement protection measures to significantly improve air quality on designated locations. This study aims in improving the ability of MLR using PCs inputs for PM10 concentrations forecasting. Daily observations for PM10 in Kuala Terengganu, Malaysia from January 2003 till December 2011 were utilized to forecast PM10 concentration levels. MLR and PCR (using PCs input) models were developed and the performance was evaluated using RMSE, NAE and IA. Results revealed that PCR performed better than MLR due to the implementation of PCA which reduce intricacy and eliminate data multi-collinearity.