The ANNIGMA-wrapper approach to fast feature selection for neural nets.

Area of Science:

• Machine Learning
• Artificial Intelligence
• Data Science

Background:

• Traditional wrapper models for feature selection are computationally expensive for large-scale neural network training.
• Backpropagation neural networks (NNs) require efficient methods for selecting relevant features from high-dimensional datasets.

Purpose of the Study:

• To develop a novel, computationally feasible feature selection approach for backpropagation neural networks.
• To improve the efficiency and accuracy of neural network models through optimized feature selection.

Main Methods:

• Introduced artificial neural net input gain measurement approximation (ANNIGMA), a weight analysis-based heuristic.
• Integrated ANNIGMA into a wrapper model framework to guide feature selection for neural networks.
• Evaluated the approach on standard datasets and real-world applications.

Main Results:

• ANNIGMA significantly reduces the number of features required for neural network training.
• The proposed method maintains or enhances model accuracy compared to existing approaches.
• Demonstrated successful application in helicopter maintenance, showcasing practical utility.

Conclusions:

• The ANNIGMA-based feature selection method offers an efficient and effective solution for neural network applications.
• This approach addresses the scalability limitations of traditional wrapper methods for large datasets.
• The technique shows promise for improving performance in complex domains such as predictive maintenance.