Sensor (group feature) selection with controlled redundancy in a connectionist framework.

Area of Science:

• Machine Learning
• Data Science
• Signal Processing

Background:

• Reducing sensor numbers is crucial for efficient decision-making in various applications.
• Sensor selection is a generalized feature selection problem, often leading to redundant feature groups.

Purpose of the Study:

• To develop a sensor selection scheme that minimizes processing time and costs.
• To propose a method for controlling redundancy among selected sensor groups.
• To enhance learning schemes for improved system performance.

Main Methods:

• A sensor (group-feature) selection scheme based on Multi-Layered Perceptron Networks was developed.
• Measures of sensor dependency were defined to quantify redundancy.
• An alternative, more effective learning scheme was presented and adapted for Radial Basis Function (RBF) networks.

Main Results:

• The proposed scheme effectively selects useful sensor groups while controlling redundancy.
• The method demonstrates effectiveness across multiple datasets.
• The new learning scheme proved more effective than the previous one.

Conclusions:

• The developed sensor selection scheme efficiently reduces processing time and costs.
• The ability to control redundancy and exploit non-linear interactions offers significant advantages.
• The scheme simultaneously selects useful groups and learns the underlying system, applicable to various learning algorithms.