SUPERVISED MACHINE LEARNING: A COMPARISON OF POISSON AND NEGATIVE BINOMIAL REGRESSION FOR COUNT DATA ANALYSIS Article Swipe
This study explores the application of supervised machine learning techniques, specifically Poisson and negative binomial regression models, for analyzing count data to forecast outgoing mail volume for the General Directorate of Posts of Saudi Arabia from 2002 to 2006. The dataset covers 13 administrative regions and consists of 65 observations with 3 variables - the dependent variable is the number of outgoing mails, and the independent variables are year and region. Exploratory data analysis revealed significant overdispersion in the data, with a large number of zero observations. Initial Poisson regression analysis highlighted the model’s limitations in addressing these data characteristics. In contrast, the negative binomial regression model demonstrated superior performance, achieving a lower Mean Absolute Prediction Error (MAPE) of 34,026.7 compared to 34,253.08 for the Poisson model. Additionally, likelihood-based metrics such as the Likelihood Ratio Test, AIC, and BIC consistently indicated that the negative binomial regression model provided a better fit to the data, reflecting the underlying overdispersion. Based on these findings, the negative binomial regression model is recommended as the primary approach for predicting outgoing mail volume for the General Directorate of Posts of Saudi Arabia.
Related Topics To Compare & Contrast
- Type
- article
- Language
- en
- Landing Page
- https://doi.org/10.17654/0972361725040
- https://pphmjopenaccess.com/index.php/aas/article/download/2355/1581
- OA Status
- hybrid
- Cited By
- 1
- Related Works
- 10
- OpenAlex ID
- https://openalex.org/W4410452994