Performance Evaluation of Levenberg-Marquardt Technique in Error Reduction for Diabetes Condition Classification

Abstract

This paper aims to provide a case study to classify diabetes medical condition amongst patients. The study examines the performance of the Levenberg-Marquardt (LM) algorithm on a single dataset, the Pima Indian Diabetes dataset, attempting to minimize error in classifying the patients as diabetes positive or negative. The learning algorithm is applied on dynamically constructed neural network to minimize the error by continuously training the network until the optimum efficiency level is obtained. The performance of the approach is verified by performing a comparison study. The comparison study involves testing of the dynamically constructed network and presents a critical analysis of the classification output. The performance of the network is measured in terms of sensitivity and specificity for different learning algorithms. The study reveals that the LM algorithm outperforms other techniques in these tests and consequently concludes it to be the best ANN learning rule in providing optimum output results when applied to a dynamically constructed neural network.