Estimate turbidity parameter in water pollution using data mining methods: Sinop province example

In this master's study, data mining models (artificial neural networks (ANN), deep learning (AN) and random forest (RO)) were used to determine turbidity, a parameter in water pollution, in various neighborhoods and seasons in the center of Sinop Province (the amount of total dissolved solids in water (TDS)) values were estimated.for this purpose, water samples were first taken in the determined locations (Osmaniye, Gelincik, Kaleyazısı, Camikebir, İncedayı, Kefevi, Ada and Zeytinlik) and in 4 different seasons, and the TDS values of these samples were determined. A total of 320 measurements were made, 10 repetitions from each group. At the same time, pH and temperature values of water samples were determined. Then, these TDS values were statistically analyzed and estimated with ANN, DL and RO models.as a result of the statistical analysis, it was determined that the season had an effect on the TDS values, but the neighborhood did not as a result of the modeling study, it was determined that all three models used were extremely successful in predicting TDS values and that these three models could be used to predict TDS values. Among these three models, the random forest model was found to be the model that gave the best performance values (accuracy (test phase) 98.92% and correlation coefficient (R2) test phase 0.966 and training phase 0.977). In the random forest model, it was determined that the highest weight among the factors was temperature (0.490) and the lowest weight was season (0.025). There are a limited number of studies on predicting environmental pollution using data mining models. Contribution to the literature can be made by increasing the studies in this field. Thus, useful information can be given to those who will work in this field.