A hybrid investment class rating model using SVD, PSO & multi-class SVM

ABSTRACT Investment class rating using machine learning techniques has received considerable research attention in the literature. Significant amount of research have been made in using SVM for binary classification problems. In recent years, support vector machine (SVM) has been attempted as a good machine learning approach for investment class rating which is a multiclass classification problem. However obtaining good classification accuracy is a challenge. Keeping this in mind, this paper proposes a hybrid investment class rating model that uses SVM for multi-class problem. The proposed model takes twelve financial ratios as attributes from different standard investment companies as inputs and correctly rates different classes as output. The suggested model works as follows: Firstly after data preprocessing, six different companding techniques have been applied to get relevant data points by removing the effect of outliers and to provide a better data fit for the SVM model for achieving better classification accuracy. Then, singular value decomposition (SVD) dimensional reduction technique is used for transforming correlated variables into a set of uncorrelated ones to better expose the various relationships among the original data items. Then, SVM model is trained offline. The overall performance of SVM strongly depends on the regularization parameter C and kernel parameter σ. So finally, we have applied the particle swarm optimization (PSO) technique using mean square error (MSE) as the fitness function to optimize the value of C and σ. The proposed scheme is implemented using Matlab and Libsvm tool. Comparison is made in terms of different performance measures like classification accuracy, sensitivity, specificity, precision etc. From experimental results and analysis, it is observed that the proposed hybrid scheme has a superior performance as compared to traditional SVM based and neural network based schemes.