Models for analysis of water suitability

Abstract

The problem of unsuitability of available drinking water for safe consumption is considered. It is proposed to use models built by machine learning methods so that when analyzing water samples it is possible to focus on the main parameters so that limited resources are not directed unnecessarily to less important features. Three methods were used when building the models: Decision Tree Classifier, Naive Bayes, and K-Nearest Neighbors. Each of these methods provides an opportunity to improve the model by selecting parameters, which was done. Three models are implemented: the DTC model using entropy when splitting a node, the Naive Bayes model with the best value of var_smoothing, and the KNN model with the optimal number of "neighbors". To evaluate the effectiveness of the proposed models, test data that were not used to build the models and several different criteria for evaluating the quality of the models were used. Ref. 5, pic. 11

Keywords

• machine learning
• analysis of the influence of factors
• classification problem