David Blumenstiel, Bonnie Cooper, Robert Welk, Leo Yi
"2021-12-10"
Project Goal: apply dimensionality reduction techniques to the Fashion MNIST dataset and evaluate the effectiveness of the results for classification using a variety of machine learning algorithms
Not all pixels are informative
We trained multiple machine learning models for classification on the engineered feature set:
We found radial SVM performed best. However, Random Forest had comparable performance and trained much faster
| Model Type | Training Duration | Test Accuracy |
|---|---|---|
| SVM (radial) | 19.0 | 0.856 |
| Mult. Log. Reg. | 9.3 | 0.824 |
| Random Forest | 5.5 | 0.855 |
| kNN | 2.6 | 0.795 |
| Naive Bayes | 0.2 | 0.713 |

