|
|
Vlpyv deliaceho bodu na klasifikáciu bankrotných firiem pomocou logitového modelu
Dolnáková, Paulína
Dolnáková, P. Impact of the threshold on the classification of bankruptcy companies using the logit model. Master thesis. Brno: Mendel University, 2021. The thesis is dedicated to creating and optimization of a predictive bankruptcy logit model. The model is based on 3 840 companies from Central Europe, among which 47 went bankrupt in 2017. To compare the quality of classification, the individual models are created both on the whole data set, but also using the division of data into training and validation subsets. We used the proportion 60/40 in which 60% of companies form a training subset and the remaining 40% of companies form a validation subset. Also, a 70/30 proportion was used. In the case of dividing the whole data set into training and validation subsets, a one thousand simulations were created. Using these strategies, separate models are created for individual monitored years, based on data from one to three years before the moment of possible bankruptcy. The quality of the classification of created models is assessed by total accuracy, sensitivity, and the AUC criterion. The influence of the threshold value on the quality of the classification is clearly presented on the individual models.
|
guest ::
