Improved Accuracy in Predicting the Best Sensor Fusion Architecture for Multiple Domains

Area of Science:

• Engineering
• Computer Science
• Data Science

Background:

• Multi-sensor fusion enhances reliability and compensates for individual sensor limitations.
• Selecting the optimal fusion architecture is complex and often relies on trial and error.
• Existing methods for predicting fusion architectures face challenges with varying dataset dimensions.

Purpose of the Study:

• To develop a novel approach for predicting the best sensor fusion architecture for a given dataset.
• To overcome limitations of previous methods in handling datasets with different numbers of variables.
• To improve the accuracy of selecting appropriate sensor fusion strategies.

Main Methods:

• Constructing a meta-dataset by extracting statistical characteristics from original datasets.
• Utilizing Sequential Forward Floating Selection for feature reduction.
• Employing a T-transform to standardize the number of features in the meta-dataset.
• Comparing the proposed method against principal component analysis-based approaches.

Main Results:

• The proposed method demonstrates improved accuracy in predicting the best sensor fusion architecture.
• The new approach effectively handles datasets with varying numbers of variables.
• Successful application across multiple domains indicates robustness.

Conclusions:

• The novel meta-dataset construction using Sequential Forward Floating Selection and T-transform is effective for predicting sensor fusion architectures.
• This approach offers a more reliable alternative to trial-and-error selection.
• The findings have significant implications for optimizing multi-sensor systems in various applications.