a study and analysis of prediction for financial crisis based on optimal feature subset selection through hybrid optimization approach

Abstract

Financial Crisis Prediction (FCP) is a multi-objective problem that determines the newlinefailing and unfailing nature of financial organizations which will predict the newlineBankruptcy at the earliest stage possible with maximum accuracy. Machine learning newlinemodels endures various phases for predicting the FCP amongst which Feature Subset newlineSelection (FSS) is significant in predicting the best features that will provide optimal newlinesolutions for FCP. The major objective of this research study is to develop an Optimal newlineFeature Subset Selection Model using Hybrid Optimization Approach that is capable newlineof modeling and recognizing best features that would be significant in predicting the newlineoutcomes of a Financial Crisis Prediction with enhanced accuracy. To test the newlineenhancement, the Financial crisis dataset was tested with machine learning algorithms newlinelike Decision Tree, PART and Naive Bayesian for classification based features, newlineLogistic Regression, Multilayer Perceptron and Radial Basis Function for newlineidentification of hidden features and AdaboostM1 for enhancing the weak features newlinealong with novel hybrid algorithm designed for prediction called Hybrid Unified newlineMachine Classifier (HUMC) which formed a combined model of classification, hidden newlinenodes identification and enhancement of weak classifiers respectively. The entire newlineresearch work comprised of three major Phases. newlineDuring the First Phase, the benchmark datasets Weislaw and Australian dataset are newlineused for initial experimentation with all machine learning algorithms viz Decision newlineTree, PART, Naive Bayesian, Logistic Regression, Multilayer Perceptron, Radial newlineBasis Function, AdaboostM1 and HUMC models. The confusion matrix is obtained newlinefrom the experiment and assessed for performance measures determined using newlineAccuracy, Sensitivity, Specificity, F-Score, Kappa Statistics etc newline