A Novel Improvement of Feature Selection for Dynamic Hand Gesture Identification Based on Double Machine Learning.

Area of Science:

• Causal inference
• Machine learning
• Data science

Background:

• Traditional machine learning and deep learning models prioritize prediction and pattern recognition.
• Causal machine learning extends these capabilities by uncovering causal relationships within data.
• Feature selection in machine learning often relies on correlational, rather than causal, links.

Purpose of the Study:

• To introduce and evaluate Double Machine Learning (DML) for causal feature selection in gesture identification.
• To demonstrate DML's ability to identify variables causally linked to gesture outcomes.
• To enhance the efficiency, performance, and interpretability of gesture classification models.

Main Methods:

• Utilizing Double Machine Learning (DML) for feature selection in a gesture identification task.
• Comparing DML-based feature selection against traditional methods like Variance Threshold, Select From Model, PCA, LASSO, ANN, and TabNet.
• Assessing the performance of classification models trained on DML-selected features.

Main Results:

• Variables selected via DML demonstrated superior performance across various classification models.
• The DML approach yielded significantly better results compared to traditional feature selection techniques.
• Causally significant variables proved more informative than correlational ones, boosting prediction accuracy and reliability.

Conclusions:

• Double Machine Learning offers a novel and effective approach for causal feature selection.
• This method enhances model interpretability and performance by focusing on causal relationships.
• DML provides a valuable perspective for building more robust and reliable machine learning models, particularly in complex datasets.