Use of Gaussian mixed regression for analysis of normally distributed data

Gaussian mixture regression models could estimate population heterogeneity through a finite set of latent classes. These models specifically focus on the identification of differences in the effect of independent variables on a dependent variable. In this thesis, a mixed regression model was established on a dependent variable such as FEV1 / FVC which is a result of pulmonary function test (PFT). FEV1 / FVC is the most important PFT parameter used to differentiate obstructive and restrictive diseases of the lung. Age, gender, height and weight were chosen as independent variables. Based on the AIC and BIC values, it was determined that the dataset was a mixture of 5 subgroups. The mean values of FEV1 / FVC in these subgroups were found to be the smallest in the first subgroup and the highest in the fifth subgroup. A regular and meaningful transition from obstructive pattern to restrictive pattern was observed by looking at FEV1 / FVC values growing from the first subgroup to the last subgroup. The first subgroup was found to be the highest in terms of weight variable but the lowest for the FEV1 / FVC variable. Therefore, the subgroup with the most obstructive pattern is the first subgroup. This result is an indication of the relationship between COPD and obesity. On the other hand, the fifth subgroup was the lowest in terms of weight, but the highest for the FEV1 / FVC variable, indicating that the weight independent variable plays an important role when it is filled in the results of the PFT. The results of this thesis showed that the use of mixture regression models in the analysis of data with a heterogeneous structure is an effective way to make sense of the data.