IMPACT OF BOX-COX TRANSFORMATION TECHNIQUE ON THE BAYESIAN MAXIMUM ENTROPY (BME) PREDICTION ACCURACY

This study investigated whether increasing the normality of an attribute using Box-Cox transformation improves Bayesian Maximum Entropy (BME) prediction accuracy. Furthermore, we examined if BME accuracy is affected by sample size or spatial dependence. For hard data, the unconditional sequential approach was used to simulate symmetric data (skewness = 0) and data positively skewed (skewness: 1, 3, 6, and 9) with sample size ranging from 100 to 500 at the interval length of 50. Soft data was randomly distributed throughout a square of unit size and a width of 1.5. Data was then transformed using Box-Cox transformation. The prediction accuracy was assessed using the Mean Square Error (MSE) and bias, and transformation methods were compared using the Multivariate Analysis of Variance (MANOVA). The results showed BME accuracy is affected by transformation methods but not the sample size and the spatial dependency. However, in comparing the transformed data with the untransformed data, the MSE and bias of the untransformed data (lambda = 1) were closer to zero than the transformed data lambda 1. As a result, we concluded that BME is robust to skewness, sample size, and spatial dependency.