Integrating multiple methods to enhance medical data classification

Abstract

In medical data classification, data reduction and improving classification performance are the important issues in the current scenario. In existing medical data classification methods, initially, the medical data pre-processing is performed. After pre-processing feature selection is performed, otherwise, the process is more time consuming and has poor accuracy. Here we have proposed two algorithms for enhancing the classification performance on medical data. In first proposed method Bag of Words technique is used for better feature subset selection. Subsequently, the hybrid Fuzzy-Neural Network approach used that can handle imprecision in data while classification. This combination of feature selection technique and Fuzzy-Neural Network classifier approach gives enhanced classification accuracy. In the second proposed algorithm, we have integrated data cleaning technique to improve data quality as pre-processing technique along with bag of words and Fuzzy-Neural Network, this method performs classification on clean filtered data with appropriately reduced feature set that results in more accurate classification than the existing methods. Thus in proposed approaches we have tried to handle three issues, removing noise in data, optimal feature subset selection and handling imprecision in data. The comparative study of various medical datasets in terms of accuracy shows that the two proposed algorithms perform better as compared to existing techniques and the enhancement obtained is around 3% and 17% respectively. In addition the performance of Bag of Words feature selection method used in the proposed system is compared with two feature selection methods LSFS and SFFS.

http://bit.ly/2Gs60hO