Expert System for Performance Prediction of Anesthesia Machines

Abstract

This paper presents an Expert System for prediction of Anesthesia machines performance and inspection requirements. It consists of Artificial Neural Network and Fuzzy classifier. The system takes 12 inputs as follows: three for measured values of volume, three for measured gas concentration, three for visual inspection of the device, average utilization time of the anesthesia machine, preventive maintenance intervals and number of additional parts. For development of the system 197 samples were used. All samples were acquired during real-time study in healthcare institutions in Bosnia and Herzegovina during the period of three years. Two-layer feedforward back propagation network with 23 neurons in hidden layer and hyperbolic tangent sigmoid transfer function was trained with 158 samples. Out of 39 validation samples, the developed network was accurate in 97.44% cases. Fuzzy rules are defined according to recommendations. Validation of developed expert system was performed using 39 samples out of which expected results were obtained for 38 samples while for 1 sample false prediction of performance status of anesthesia machine was recorded.