A non-linear data mining parameter selection algorithm for continuous variables.

Area of Science:

• Data Mining
• Statistical Modeling
• Machine Learning

Background:

• Traditional regression analysis often struggles with non-linear data and identifying optimal variable subsets.
• Existing methods may not fully capture complex relationships or synergistic effects between variables.

Purpose of the Study:

• To propose a new data mining algorithm for capturing non-linearity and selecting optimal variable subsets.
• To introduce an enhanced regression analysis technique that transforms predictors and explores synergistic effects.
• To develop an efficient model selection process that yields stable and interpretable models.

Main Methods:

• The proposed algorithm integrates variable transformation and subset selection within a classical least squares regression framework.
• It employs mathematical transformations of predictors to capture complex relationships and synergistic effects.
• The method combines all possible subset selection with variable transformations and interaction exploration.

Main Results:

• The algorithm produces an optimal subset of variables, improving model selection efficiency.
• It introduces interpretable parameters through input transformation, ensuring a faithful data fit.
• The method effectively minimizes mean square error and variability, controlling multicollinearity.

Conclusions:

• The developed data mining algorithm offers an efficient and stable approach to regression modeling.
• It successfully addresses non-linearity and complex variable interactions, leading to optimal explanatory variable sets.
• This technique enhances the classical least squares regression framework with automatic variable transformation and model selection.