Performance improvement in a classification method by using statistical pre-processing

One of the most significant current discussion in AI (Artificial Intelligence) and HCI (Human Computer Interface) is the pattern recognition algorithm. Many methods are available for this purpose, such as, the support vector machine, artificial neural network, and Bayesian decision rule. In these methods, the number of features is the most critical factor affecting the classifier performance. Therefore, we herein propose feature selection and extraction methods to obtain a more effective classifier (higher accuracy and less complexity). To do this, we apply a statistical algorithm. Before we use pattern recognition algorithms, we select features using variance and correlation coefficient. Additionally, we extract the features using the dimension reduction method. We could filter out critical features and reduce the number of features using above process. For an objective evaluation, we use electroencephalogram and the survey data of the DEAP (dataset for emotion analysis using physiological signals). Additionally, we perform a comparison with the existing study. According to the performance evaluation, a classifier with higher accuracy and less computational complexity is obtained. © ICROS 2019.