Fig. 5: Results Comparison of PDD and other ML Models.
From: Theory and rationale of interpretable all-in-one pattern discovery and disentanglement system
In entity clustering, PDD outperformed K-Means (N) and K-Means (D) before and after error correction in recall (blue bar), precision (red bar), F-measure (gray bar), and accuracy (the orange line). In Class Association, note that the improved results in other ML models represented by the red bars were made possible when the outliers identified by PDD were removed. Without that, their accuracies were represented only by the blue bars, while those of PDD were represented by the red bars. The results show that in class association, PDD outperformed existing popular ML models significantly after class readjustment. Even before class readjustment, its class association results are comparable to those obtained by the existing supervised models. For the Classification of Imbalanced Classes, PDD (orange bars) outperformed other methods such as Logistic Regression, CART (Decision Tree), and Naive Bayes significantly.
