Abstract

This paper presents a simulation-based performance evaluation of global-local hybrid ensemble (GLHE) in comparison to hundreds of classifiers which competed in the KDD 2004 Cup for the particle physics task. Simulations are performed on Weka machine learning software workbench. Performance metrics include prediction accuracy, area under the receiver operating characteristic (ROC) curve, cross entropy, and Q-score. Simulation results indicate that the difference between the performances of GLHE and the best submission for the KDD Cup 2004 particle physics task is relatively small for both prediction accuracy and area under the ROC curve. The performance of GLHE for prediction accuracy is only 3.2% worse than that of the best submission. On average over the four metrics, GLHE is ranked 56.5 out of 109. Noting that GLHE has not been fine-tuned or optimized with respect to any adjustable parameters or otherwise while the competition entries most likely were, the results are promising in support of the claims for the robustness of GLHE performance.