A Collimator Neural Network Model for the Classification of Genetic Data

It is extremely challenging to design a machine learning algorithm that is able to generate tolerable error rates for the class prediction of various types of genetic data. A particular system may be very effective for one microarray dataset but fail to perform in a similar fashion for another dataset. This paper introduces a neural approach to use Generalized Regression Neural Network (GRNN), Collimator Neural Network (CNN) which provides consistent performance stability for various types of microarray data. CNN performance has been cross validated with the k-fold cross validation (leave one out) technique for BRCA1, BRCA2 and Sporadic mutation classification for ovarian and breast cancer data. The paper presents comparative classification results for binary and multiclass prediction by CNN against previously presented work and other well established classifiers including, Support Vector Machines (SVM), Probabilistic Neural Network (PNN), single GRNN and novel Concurrent PNN (CPNN). Simulation results confirm that CNN performed significantly better than SVM, PNN, GRNN and CPNN for both the breast cancer and ovarian cancer datasets