Data Mining of 24-Hour Cumulative Precipitation Data in Iran Using Machine Learning: Multilayer Perceptron Neural Network and Decision Tree

Research Topic: This study examines 24-hour cumulative precipitation data in Iran and predicts rainfall amounts over various time periods using data mining and machine learning. Objective: To develop an accurate model for predicting 24-hour cumulative precipitation in regions of Iran using multilayer neural networks and decision trees to improve hydrological planning and water resource management. Method: A daily precipitation dataset D was collected from Iranian stations and prepared using normalization. Two machine learning models including MLP with activation function σ and decision tree with entropy criterion were implemented. The models’ performance was evaluated and compared with accuracy, precision, and error criteria. Results: The MLP model demonstrated efficiency in estimating monthly precipitation by minimizing MSE to 0.04. The decision tree analysis classified Iran provinces into seven clusters based on precipitation characteristics; clusters 4 and 7 represent provinces with minimum (including Isfahan, Sistan and Baluchestan, Yazd) and maximum precipitation (including Gilan, Kohgiluyeh and Boyer-Ahmad, Mazandaran), respectively. Linear regression showed a significant effect of the time variable with 0.209 on precipitation variance. Conclusions: The use of machine learning, especially neural networks, is effective in analyzing hydrological data in Iran and can help improve precipitation forecasting systems.