Investigating the reliability of a low-back-pain MLP by using a full explanation facility

This study investigates the reliability of a low-back-pain multilayer perceptron network from a hidden layer decision region perspective. Using decision region information from an explanation facility the training examples are discovered to occupy decision regions in contiguous class threads across the 48-dimensional input space. Test cases show a similar distribution and consistency within the contiguous threads but with a reduced reliability. Three test regions outside the network´s knowledge bounds are situated between training regions with a consistent classification. The hypothesis that classifications are reliable within the knowledge bounds and potentially unreliable outside the knowledge bounds is examined