Effects of Parametric and Non-Parametric Methods on High Dimensional Sparse Matrix Representations

This repository is the implementation of Effects of Parametric and Non-Parametric Methods on High Dimensional Sparse Matrix Representations. If you use this repo or paper, please consider citing it.

The dataset used can be found here.

Citation :

@article{tambe2022effects,
  title={Effects of Parametric and Non-Parametric Methods on High Dimensional Sparse Matrix Representations},
  author={Tambe, Sayali and Joshi, Raunak and Gupta, Abhishek and Kanvinde, Nandan and Chitre, Vidya},
  journal={arXiv preprint arXiv:2202.02894},
  year={2022}
}

Results :

The results are calculated on the basis of classification metrics. Precision, Recall and F1-Score.

Precision :

The precision for 50 dimensions of the representations over used algorithms is checked.

Algorithm	Macro Average	Weighted Average
Linear Discriminant Analysis	89%	90%
Naive Bayes	79%	80%
Decision Tree	89%	90%
Support Vector Machine	90%	91%

Similarly Macro Averaging Precision is calculated for 100, 500, 1000, 5000 dimensions of sparse matrix representations.

Dimensions	LDA Naive	Bayes	Decision	Tree SVM
100	90%	83%	89%	92%
500	91%	85%	91%	93%
1000	92%	86%	91%	93%
5000	92%	86%	92%	93%

Similarly Weighted Averaging Precision is calculated for 100, 500, 1000, 5000 dimensions of sparse matrix representations.

Dimensions	LDA	Naive Bayes	Decision Tree	SVM
100	91%	83%	90%	92%
500	92%	85%	91%	93%
1000	92%	86%	91%	93%
5000	92%	86%	92%	94%

Recall :

The Macro Averaging Recall for 50, 100, 500, 1000, 5000 dimensions of sparse matrix representations.

Dimensions	LDA	Naive Bayes	Decision Tree	SVM
50	88%	79%	88%	90%
100	89%	82%	89%	91%
500	91%	84%	91%	93%
1000	91%	85%	91%	93%
5000	91%	85%	92%	93%

The Weighted Averaging Recall for 50, 100, 500, 1000, 5000 dimensions of sparse matrix representations.

Dimensions	LDA	Naive Bayes	Decision Tree	SVM
50	88%	80%	89%	90%
100	89%	84%	89%	91%
500	91%	85%	91%	93%
1000	91%	86%	91%	93%
5000	92%	87%	92%	93%

F1-Score :

F1-Score for 50, 100, 500, 1000, 5000 dimensions of sparse matrix representations.

Dimensions	LDA	Naive Bayes	Decision Tree	SVM
50	88%	80%	89%	90%
100	89%	84%	89%	91%
500	91%	85%	91%	93%
1000	91%	86%	91%	93%
5000	92%	87%	92%	93%