Overfit Analysis on Decision Tree Classifier for Fault Classification in DAMADICS

In this paper, the effect of overfitting displayed by a decision tree classifier model is studied and the method of resampling technique to eliminate the overfitting is implemented in the pre-pruning stage of the algorithm. The classifier is built for fault classification function subjected to a synthetically generated dataset of the benchmark DAMADICS process which represents a pneumatic actuator system. The overfitting problem for both multiclass classification and binary class classification using maximum depth as the optimized hyper parameter is analyzed. The results before and after eradicating the overfit are tabulated. The performance of the model is plotted between hyper parameter chosen and the testing, training accuracy. The best fit tree model is also graphically visualized.