E2E-FS: An End-to-End Feature Selection Method for Neural Networks.

Area of Science:

• Machine Learning
• Data Science
• Artificial Intelligence

Background:

• Traditional embedded feature selection methods, such as tree-based algorithms and LASSO variants, offer either explainability or accuracy, but not both.
• Tree-based methods excel at explaining variable importance, while LASSO variants prioritize predictive performance over detailed explanations.

Purpose of the Study:

• To present a novel embedded feature selection algorithm, End-to-End Feature Selection (E2E-FS), designed to achieve both high accuracy and enhanced explainability.
• To address the limitations of existing methods by integrating a mechanism for explicit feature selection within the model training process.

Main Methods:

• Developed End-to-End Feature Selection (E2E-FS), a novel embedded feature selection algorithm.
• Utilized gradient descent techniques for optimization, incorporating specific restrictions to enforce the selection of a maximum number of features.
• Employed non-convex regularization terms within the algorithm's framework.

Main Results:

• Experimental results demonstrate that E2E-FS effectively provides both accuracy and explainability.
• The algorithm successfully selects a constrained set of features crucial for subsequent classification tasks.
• E2E-FS proved compatible with various learning models trained via gradient descent.

Conclusions:

• End-to-End Feature Selection (E2E-FS) offers a promising approach to embedded feature selection, merging the strengths of explainability and accuracy.
• The algorithm's design allows for its application across a wide range of gradient descent-based machine learning models.
• E2E-FS represents a significant advancement in developing more transparent and effective feature selection techniques.