The Point Nuisance Method as a decision-support system based on bayesian inference approach

The article attempts to transfer information from the Point Nuisance Method (PNM) used in Poland in the issue of protection of buildings in mining areas, to the system of inference based on Bayesian formalism. For this purpose, all possible combinations occurring in PNM were selected. The number of numerically generated patterns was 6,718,464 cases. Then, based on Python package Scikit-Learn, a classification model was created in the form of the Naïve Bayes Classifier (NBC). The effectiveness of three methods used to build this type of decision-support system was analysed, from which the Categorical Multinomial Naive Bayes (CMNB) approach was finally selected. With the created classifier, its properties were verified in terms of quality of classify and generalization. For this purpose a general approach was used, analysing the level of accuracy of the model in relation to training and teaching data, and detailed, based on the analysis of the confusion matrix. Additionally, the operation of the created classifier was simulated to determine the optimal Laplace smoothing parameter α. The article ends with conclusions from the carried out calculations, in which an attempt was made to answer the question concerning potential reasons for incorrect classification of the created CMNB model. The discussion ends with a reference to the planned research, in which, among other things, the use of more complex Bayesian belief networks (BBN) is planned.