Friedman Test

A Statistical Comparisons of Classifiers over Multiple Data Sets

Introduction

The objective of this assignment is to perform a robust statistical comparison of multiple machine learning classifiers across several independent datasets. When evaluating more than two models over multiple datasets, traditional parametric tests such as the Analysis of Variance (ANOVA) are often unsuitable. This is because the assumptions of a normal distribution and homogeneity of variance are frequently violated by classification accuracy scores. To address this, we employ the Friedman test, a non-parametric (distribution-free) alternative, which rather than analyzing raw performance scores, operates on the relative ranks of the algorithms.

Nemenyi Post-Hoc Test

Datasets

Visual representation of the standardized datasets.

As shown in ,

Models

Evaluation Protocol

Results and Discussion

Look at the results in .

Loading table...

Mean classification accuracy scores for different classifiers...

Critical Difference

Critical Difference diagram.

is the Critical Difference diagram

Conclusion

Visual representation of the decision boundaries generated by the different classifiers over the datasets.

According to research , notes are vital.

---

References

Loading bibliography...