A Novel Performance Metric for Building an Optimized Classifier
Mohammad Hossin, Md Nasir Sulaiman, Aida Mustapha and Norwati Mustapha
DOI : 10.3844/jcssp.2011.582.590
Journal of Computer Science
Volume 7, Issue 4
Problem statement: Typically, the accuracy metric is often applied for optimizing the heuristic or stochastic classification models. However, the use of accuracy metric might lead the searching process to the sub-optimal solutions due to its less discriminating values and it is also not robust to the changes of class distribution. Approach: To solve these detrimental effects, we propose a novel performance metric which combines the beneficial properties of accuracy metric with the extended recall and precision metrics. We call this new performance metric as Optimized Accuracy with Recall-Precision (OARP). Results: In this study, we demonstrate that the OARP metric is theoretically better than the accuracy metric using four generated examples. We also demonstrate empirically that a naïve stochastic classification algorithm, which is Monte Carlo Sampling (MCS) algorithm trained with the OARP metric, is able to obtain better predictive results than the one trained with the conventional accuracy metric. Additionally, the t-test analysis also shows a clear advantage of the MCS model trained with the OARP metric over the accuracy metric alone for all binary data sets. Conclusion: The experiments have proved that the OARP metric leads stochastic classifiers such as the MCS towards a better training model, which in turn will improve the predictive results of any heuristic or stochastic classification models.
© 2011 Mohammad Hossin, Md Nasir Sulaiman, Aida Mustapha and Norwati Mustapha. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.