Akademik Veri Yönetim Sistemi
6th INTERNATIONAL CONGRESS ON ENGINEERING AND LIFE SCIENCE, Girne, Kıbrıs (Kktc), 2 Eylül - 04 Ekim 2025, ss.23-32, (Tam Metin Bildiri)
Drought is one of the most significant natural disasters, causing severe negative impacts on agriculture, water resources, and ecosystems. Strategies for disaster risk reduction, agricultural planning, and water management all depend on early and precise drought prediction. Palmer Drought Severity Index (PDSI) data from Kayseri province for the years January 1948–December 2023 were used in this study to anticipate drought. Model training was conducted using data from January 1948 to March 2007, and the test dataset was defined as the time frame from April 2007 to December 2023. PDSI values from time lags t-1, t-2, t-3, and t-4 were used as input variables, and the drought level at time t was chosen as the output variable. Support Vector Regression and Gaussian Process Regression (GPR) algorithms were used for the prediction of the PDSI. Tunable Q-Factor Wavelet Transform (TQWT) and Variational Mode Decomposition (VMD) were employed as preprocessing techniques to develop hybrid models. The Coefficient of Determination (R2 ), Mean Absolute Error (MAE), and Root Mean Square Error (RMSE) metrics were used to assess the four developed hybrid models’ performance. Based on the evaluations conducted on the test data, the TQWT-GPR model demonstrated the best performance among the four developed hybrid models. The TQWT-GPR model was able to predict PDSI values with performance values of MAE = 0.061, RMSE = 0.075, and R2 = 0.999. The results indicate that the GPR model exhibited greater compatibility with the applied preprocessing techniques. For further improvement in forecasting performance, particularly in MAE and MSE, the integration of advanced techniques such as reinforcement learning is recommended. Keywords: Palmer Drought Severity Index, Machine Learning, Prediction, Türkiye.