The Use of Hybrid Information Retrieve Technique and Bayesian Relevance Feedback Classification on Clinical Dataset

Abstract

Retrieval of information related to a subset variable or feature has become the attention of many researchers in data mining fields. The objective of feature selection (FS) is to improve the performance of the prediction. This contributes to providing a better definition of the features, feature structure, feature ranking, feature selection functions, efficient search techniques, and feature validation methods. In this study, a retrieval method that integrates correlation and linear forward selection algorithms to evaluate and generate the subset of clinical features are present. The objective of the research is to find the optimal features of a cancer dataset and to classify the disease into multiple cancer stages: one, two, three, and four. The research methodology is developed based on data mining, knowledge data discovery with four phases: pre-processing, resampling, feature selection, and classification. The proposed Bayesian Relevance Feedback (BRF) for classification is also described to resolve the zero value of posterior probabilities, concentrating on increasing the accuracy in the diagnosis of cancer stages. The experimental works are done on oral cancer dataset by applying WEKA. The analysis on accuracy performance was done on several classification algorithms using 15 optimal features that were chosen by a hybrid features selection method. The result shows that, BRF has outperformed others achieving 97.25% classification accuracy compared to the six classifiers, which are K-Nearest Neighbors Classifier, Multi Class Classifier, Tree-Random, Multilayer Perceptron, Naïve Bayes, and Support Vector Machine.