THE USING OF PARTIAL LEAST SQUARE REGRESSION (PLSR) TO SOLVE THE COLINIERTY PROBLEM AND ITS UTILIZATION IN PREDICTING THE MONSOON RAINFALL , BASED ON GLOBAL CIRCULATION MODEL (GCM) DATA

Abstract

Prediction of parameter in multiple regression analysis is an interesting topic of some researches. This can be due to some problems that may rise in regression,and one of these is a colinierty problem. The method usually used to solve this problem is Principle Component Regression (PCR), Ridge Regression, and Partial Least Square (PLS). Differ from PCR and Ridge Regression, regression coefficient in the PLS is obtained iteratively and doesn’t have a closed formula to get a variant of regression coefficient. GCM data were obtained by using the grade ranged from v2272 – v2727 and the rainfall data from Kotabaru BMKG Station collected from 1996 – 2001. These data were then used as a model, whereas the data collected in 2001 was used to validate the model. All calculation and figures were made by using Minitab software 1412 version and Microsoft Excel 2003. Estimation with PLS algorithm to various iteration resulted in PRESS and R 2 values. Variant of X variable with 3 components of model could explain a 66.9% of variant of dependent variables. Minimum PRESS with the value of 5.714.118 was reached in the third iteration with the highest prediction of R 2 value as 7%. Regression analysis using PLS method based on GCM and rainfall data or is known as Statistical Downscaling, resulted in a value prediction of a coefficient regression in the third iteration. This model has been validated and can be used to predict rainfall during the ne xt 12 months. It can conclude that the optimum PLS model using GCM data can be used to predict a local climate , whereas as a
suggestion is that we still need to see the relation between rainfall variable and seasonal period, so that it can be used to predict every period of event, as in case of DJF periode (Dec, Jan, and Peb).