Feature Interaction Detection in Big Data Through a New Choquet Integral based Deep Neural Network.

Area of Science:

• Artificial Intelligence
• Machine Learning
• Data Science

Background:

• Big data analysis requires advanced algorithms for complex feature interactions.
• Standard methods often fail to capture multi-feature subset relationships.
• Identifying synergistic and antagonistic relationships is crucial for accurate modeling.

Purpose of the Study:

• To develop a novel activation function for deep neural networks to model complex interactions in high-dimensional data.
• To introduce a sub-additive tool for analyzing weighted feature compilations.
• To apply and validate the method on real-world health data for weight loss prediction.

Main Methods:

• Development of a novel Choquet Integral activation function for deep neural networks.
• Transformation of high-dimensional data into simpler sub-feature sets.
• Utilizing balanced fuzzy measures and sub-additivity principles.
• Testing and hyper-parameter optimization on health-related datasets.

Main Results:

• The Choquet Integral activation function effectively models complex interactions and non-linear dependencies.
• The method identifies sub-additive feature interactions missed by standard approaches.
• The model demonstrates robust performance on standard benchmarks using health data.

Conclusions:

• The novel activation function offers a powerful tool for big data analysis, particularly in identifying complex feature interactions.
• This approach advances the modeling of synergistic and antagonistic relationships among features.
• The method has broad applicability across diverse fields including biomedicine, finance, and cybersecurity.