|
|
 |
| |
|
|
|
Developing a New Estimation Approach for Constructing a Flexible Location Model to Address Challenges with Numerous Empty and Non-Empty Cells |
|
|
|
PP: 1095-1103 |
|
|
|
doi:10.18576/amis/180515
|
|
|
|
Author(s) |
|
|
|
Hashibah Hamid,
Nor Azrita Mohd Amin,
Saadi A. Kamaruddin,
Ayu Abdul-Rahman,
Friday Z. Okwonu,
|
|
|
|
Abstract |
|
|
| This paper aims to address the challenges posed by the simultaneous occurrence of numerous empty and many non-empty cells in the Location Model (LM). The LM is a classification method used in scenarios with mixed variables to distinguish between two groups. However, the classical LM, relying on maximum likelihood estimation (MLE), faces challenges when encountering empty cells due to its assumption that all categorical variables are binary. This assumption leads to exponentially growing cells with binary variables, increasing the likelihood of encountering empty cells, especially with numerous binary variables or small sample sizes. Although the LM applies smoothing techniques to mitigate this issue, it has limitations with many binary variables or small samples observed in the study. To tackle these troubles, this paper develops a new parameter estimation approach combining MLE and smoothing for tackling empty and non-empty cells. The outcome of this new estimation yields a flexible LM, which proficiently manages numerous binaries or limited sample sizes, thereby enhancing performance and adaptability across diverse cell scenarios; whether involving many empty or many non- empty cells. This innovative approach offers a promising solution to longstanding challenges in classification tasks, particularly in critical domains like cancer treatment selection, and sets a new standard for precise classification, empowering researchers and practitioners with enhanced decision-making tools.
|
|
|
|
|
|
|
|
