A Hybrid Prediction System Using Rough Sets and Artificial Neural Networks

This paper illustrates a hybrid prediction system consists of Rough Set Theory (RST) and Artificial Neural Network (ANN) for processing medical data. In the process of developing a new data mining technique and software to aid efficient solutions for medical data analysis, we propose a hybrid tool that incorporates RST and ANN to make efficient data analysis and suggestive predictions. In the experiments, we used spermatological data set for predicting quality of animal semen. The data set used in the experiments is subjected to quantize and normalize, and use this as a reflection of the internal system state. The RST is used as a tool for reducing and choosing the most relevant sets of internal states for predicting the semen fertilization potential. Chosen optimal data set is input to constructed neural network with supervised learning algorithm for the prediction of semen quality. This paper demonstrates that the RST is an effective pre-processing tool for reducing the number of input vector to ANN without reducing the basic knowledge of the information system in order to increase prediction accuracy of the proposed system. The resulting system is a hybrid prediction system for medical database called an Intelligent Rough Neural Network System (IRNNS).