Artificial Bee Colony for Features Selection Optimization in Increasing T-Method Accuracy

Abstract

The study of prediction has drawn great interest in a wide range of field. T-Method which was developed specifically for prediction of the multidimensional case using historical data to develop its baseline model proved that making a prediction is possible even with limited sample size. The element of the signal to noise ratio (SNR) adopted into the T-Method strengthens its robustness. Orthogonal array (OA) in T-Method was used as features selection optimization in improving the analysis speed, cost and computer burden during the analysis. However, the limitation of OA in dealing with higher dimensionality and complex combination factors restraint the optimization accuracy. Artificial Bee Colony (ABC) was adopted in this study to overcome this limitation. The result of this study shows that T-method +ABC provide the best error% accuracy with only 2.45% and 2.53% (3 optimized features out of 15) compared to T-Method +OA which 2.81% and 2.67% and T-Method +Spearman Correlation as 3.16% and 3.06%. The power consumption prediction case study is a good example for cases that deal with high correlation coefficient (R2) baseline model (>0.8). If the R2 is lower than 0.8, further enhancement needs to be done to ensure a low risk of high error% prediction. Â