On the use of fuzzy C-regression truncated models for health indicator in intensive care unit

Abstract

Two new techniques for clustering data, namely the fuzzy c-regression truncated models (FCRTM) and fuzzy c-regression least quartile difference (LQD) models (FCRLM) were proposed in this thesis in analyzing a nonlinear model. These new models include their functions, the estimation techniques and the explanation of the five procedures. The stepwise method was used for variable selection in the FCRTM and FCRLM models. The number of clusters was determined using the compactness-to-separation ratio, NEW F . The various values of constant, k (k = 0.1, 0.2, ..., 8) in generalized distance error and various values of fuzzifier, w (1< w <3) were used in order to find the lowest mean square error (MSE). Then, the data were grouped based on cluster and analyzed using truncated absolute residual (TAR) and the least quartile difference (LQD) technique. The FCRTM and FCRLM models were tested on the simulated data and these models can approximate the given nonlinear system with the highest accuracy. A case study in health indicator (simplified acute physiology score II (SAPS II score) when discharge from hospital) at the intensive care unit (ICU) ward was carried out using the FCRTM and FCRLM models as mentioned above. Eight cases of data involving six independent variables (sex, race, organ failure, comorbid disease, mechanical ventilation and SAPS II score when admitted to hospital) with different combinations of variable types in each case were considered to find the best modified data. The comparisons among the fuzzy cmeans (FCM) model, fuzzy c-regression models (FCRM), multiple linear regression model, Cox proportional-hazards model, fuzzy linear regression model (FLRM), fuzzy least squares regression model (FLSRM), new affine Takagi Sugeno fuzzy models, FCRTM models and FCRLM models were carried out to find the best model by using the mean square error (MSE), root mean square error (RMSE), mean absolute error (MAE) and mean absolute percentage error (MAPE). The results showed that the FCRTM models were found to be the best model, having the lowest MSE, RMSE, MAE and MAPE. This new modelling technique could be proposed as one of the best models in analyzing mainly a complex system. Hence, the health indicator in the ICU ward could be monitored by managing six independent variables and other management quality variables in the hospital management.