Comparative Analysis between Artificial Neural Networks and Statistical Techniques in Bioinformatics

Abstract

Researchers sometime have difficulty to determine the right data mining technique to be used as many aspects need to be considered beforehand. This paper presents an analysis of two data mining techniques namely statistical and neural networks (NNs). The analysis focused on bioinformatics domain specifically cancer data. The research seeks to compare the ability of NNs with Statistical method using NeuroSolution software. A sufficient amount of data sets was divided for the purpose of training, validation and testing. The specific techniques used for these studies were backpropagation algorithm (NNs) and multiple linear regressions (MLR). The aspects of analysis include the accuracy of generalisation, the computation complexity and the complexity of the model. The ability of these techniques in classifying the classes of benign and malignant was experimented and monitored. The results have demonstrated that NNs is better in classifying the targeted data. The means squared error (MSE) and root mean squared error (RMSE) generated from NNs were less if compared to the multiple linear regression technique. Additionally, NNs also showed higher accuracy in classifying the classes and easily handled although more input variables were used. MLR on the other hand gave low accuracy and involved tedious computation. The findings have revealed the ability of NNs in many aspects as been mentioned if compared to its counterpart.