Controlling Sensitivity of Gaussian Bayes Predictions based on Eigenvalue Thresholding

Abstract

Gaussian Bayes classifiers are widely used in machine learning for various purposes. Its special characteristic has provided a great capacity for estimating the likelihood and reliability of individual classification decision made, which has been used in many areas such as decision support assessments and risk analysis. However, Gaussian Bayes models tend to perform poorly when processing feature vectors of high dimensionality. This limitation is often resolved using dimension reduction techniques such as Principal Component Analysis. Conventional approaches on reducing dimensionalities usually rely on using a simple threshold based on accuracy measurements or sampling characteristics but rarely consider the sensitivity aspect of the prediction model created. In this paper, we have investigated the influence of eigenvalue selections on Gaussian Bayes classifiers in the context of sensitivity adjustment. Experiments based on real-life data have shown indicative and intriguing results.