Region of Interest Selection for Functional Features.

Area of Science:

• Machine Learning
• Statistical Modeling
• Data Science

Background:

• Feature selection is crucial for supervised learning but can be computationally intensive (NP-hard).
• Cross-validation, commonly used to prevent overfitting with limited data, incurs high computational costs.
• Existing methods often struggle to balance model performance and computational efficiency.

Purpose of the Study:

• To develop an innovative feature selection strategy for functional linear regression.
• To reduce the risk of overfitting without employing cross-validation.
• To identify optimal sub-intervals (regions of interest) of functional features.

Main Methods:

• Proposes a novel method for selecting optimal sub-intervals (regions of interest) in functional linear regression.
• Evaluates overfitting risk for each sub-interval by calculating the necessary sample size for statistical power.
• Combines overfitting risk assessment with model accuracy measures to rank and select the best region of interest.

Main Results:

• The proposed method effectively reduces overfitting risk without the need for cross-validation.
• Achieved excellent prediction accuracy compared to state-of-the-art feature selection techniques.
• Demonstrated substantial reductions in computational cost.

Conclusions:

• The novel feature selection approach offers a computationally efficient and accurate alternative for functional linear regression.
• Identifies optimal regions of interest within functional predictors, improving model generalization.
• Provides a valuable tool for researchers dealing with high-dimensional functional data and limited sample sizes.