Heart Diseases Detection Using Fuzzy Hyper Sphere Neural Network Classifier

This paper uses Fuzzy Hyper Sphere Neural Network (FHSNN), a supervised neural network, which utilizes fuzzy sets as pattern classes for detection of heart diseases. A class fuzzy set in FHSNN is a union of fuzzy set hyper spheres from the same class. Two standard heart databases from machine learning UCI repository has been selected for the experimentation, which are SPECTF, STATLOG (Heart) databases. For STATLOG data set, FHSNN has created 133 hyper spheres and given 52.592% recognition rate. For SPECTF data set it has created 66 hyper spheres and given 48.1283% recognition. The theoretical performance of FHSNN in terms of true error with 0.99% confidence lie in interval [0.18, 0.62], approximately. Training time is slightly more but, it needs to be trained only once so it can not be treated as a serious drawback. FHSNN is having capability to learn the patterns on fly makes it as a distinguished classifier from others. Its learning algorithm is simpler, easy to understand and can be parallelized easily. We can improve the percentage recognition rate by increasing the training set size. By observing training and recall time of FHSNN, we can say that it is suitable for real time heart diseases detection.