2003 Volume 11 Issue 2 Pages 85-91
Urinary incontinence is deeply related to the rehabilitation of brain disease patients, and it is an important matter of concern for clinical nursing. The approach of pattern recognition reported in this paper is to predict the urinary continence recovery four weeks after the surgical operation or stroke. The prediction is made based on the information including the attributes of the patients, type of brain disease, the level of consciousness (i.e., eye opening, verbal response, or motor response) and the level of continence one week after the operation or stroke. From the aspect of cost for prediction, the item selection method using orthogonal arrays was used. In the past study, items were selected in such a way to maximize the difference in Mahalanobis distance between the two groups of patients: the one, which recovered urinary continence after four weeks, and the one, which could not. Since urinary continence recovery was evaluated by an eight level score, the recovery level was omega transformed to be used as the signal, and Mahalanobis distance in logarithmic scale was used as the output. Based on such input and output, dynamic characteristic was used for item selection. This study shows how the signal factor was set and how the results were evaluated for item selection using dynamic characteristic in MT system.
