Discrimination and allocation using a mixture of discrete and continuous variables with some empty states

Area of Science:

• Statistics
• Machine Learning
• Data Science

Background:

• Classification rules are essential for data analysis.
• Real-world data often presents challenges like mixed variable types (binary and continuous).
• Data sparsity (zero or few observations in some categories) complicates rule estimation.

Purpose of the Study:

• To estimate a likelihood ratio classification rule based on location mode.
• To handle datasets with both binary and continuous variables.
• To address the practical issue of data sparsity in classification.

Main Methods:

• Utilized iterative proportional fitting for log-linear model approximation.
• Employed a linear additive model for parameter estimation.
• Assessed the performance using estimated error rates.

Main Results:

• Successfully estimated a classification rule despite data sparsity.
• Demonstrated a viable method for handling mixed data types.
• Evaluated the rule's effectiveness through error rate estimation.

Conclusions:

• The proposed method provides a robust approach for classification with mixed data and sparsity.
• Iterative proportional fitting and linear additive models are effective for estimating such rules.
• The approach is suitable for practical applications where data limitations are common.