Advances and Applications in Statistics
Volume 55, Issue 1, Pages 67 - 76
(March 2019) http://dx.doi.org/10.17654/AS055010067 |
|
A COMPARISON OF LOGISTIC REGRESSION AND MACHINE LEARNING ALGORITHMS APPLIED TO ZERO COUNTS DATA IN CONTINGENCY TABLES
Nurin Dureh and Phattrawan Tongkumchum
|
Abstract: Logistic regression is widely known to not converge when applied to zero count data in a contingency table. With the emergence of non-parametric methods, especially machine learning techniques for data analysis, it is interesting to see if these techniques can be adapted to such a problem. This study provides a comparison of the predictive accuracies among the logistic regression model and select machine learning methods, namely recursive partitioning, neural networks, and random forest, specifically applied to data with zero counts. The data modification method (DM) was also applied before logistic regression. The results revealed that logistic regression with the DM method performs had predictive accuracy equal to random forest or recursive partitioning. The neural network had weak performance both with real and with simulated data. |
Keywords and phrases: logistic regression, data modification, random forest, neural network, recursive partitioning.
|
|
Number of Downloads: 470 | Number of Views: 3677 |
|