Design and development of decision making framework for small dataset

Abstract

Decision Making (DM) process becomes complicated when there are too many features or when the number of records for analysis is small. Dimensionality reduction techniques in conjunction with classification have gradually emerged as a prevailing theory for decision making. However, Feature extraction techniques like Principal component analysis and Independent component analysis are employed to reduce the dimension but the underlying feature which influences the decision is unknown. Feature Subset Selection algorithm reduces the dimension and determines the decision variables but Feature Subset Selection evaluated by wrapper method requires large dataset. A Decision Making framework is proposed to address these issues. The proposed decision-making framework incorporates four components namely (1) Feature Selection, and (2) Classification. The feature selection process newlineincludes two methodologies (1) Feature Subset Selection and (2) Feature Extraction. Feature subset selection intend to select the most relevant and nonredund ant features, where redundant data are removed by forming correlation based maximum spanning clusters and the most relevant data is chosen using fisher score from each cluster. Feature Extraction technique is employed on the eliminated features to improvise the classification accuracy as an ensemble framework. Linear Principal Component Analysis, Nonlinear Principal analysis and Independent component analysis are analyzed as a wrapper method to choose the best transformation technique. Multiple hybrid kernels are introduced in Support Vector Machines to improve classification accuracy for small data effectively. newline newline