MONTHLY RAINFALL TIME SERIES PREDICTION IN AMASYA USING SHUFFLED FROG-LEAPING ALGORITHM AND SOFTCOMPUTING


Katipoğlu O. M.

16th INTERNATIONAL ISTANBUL SCIENTIFIC RESEARCH CONGRESS IN SCIENCE, ENGINEERING, ARCHITECTURE AND MATHEMATICS, İstanbul, Türkiye, 28 - 29 Şubat 2024, ss.1

  • Yayın Türü: Bildiri / Özet Bildiri
  • Basıldığı Şehir: İstanbul
  • Basıldığı Ülke: Türkiye
  • Sayfa Sayıları: ss.1
  • Erzincan Binali Yıldırım Üniversitesi Adresli: Evet

Özet

Metahiuristic optimization is an optimization approach inspired by the behavior of living things

in nature. Using default parameters in artificial intelligence (AI) models may prevent prediction

performance from reaching the best values with local optimum values instead of global optimum results.

Therefore, hyperparameters must be used to produce precise predictions with AI models. In this study,

the prediction of monthly average precipitation was combined with the metaheuristic Shuffled Frog-

Leaping Algorithm (SFLA) and artificial neural network (ANN). In creating the model, rainfall data in

Amasya was estimated by using rainfall data from neighboring stations such as Çorum, Tokat and Samsun

as input. The effect of Resilient Backpropagation and Model Gradient Descent activation functions in

the setup of the models was also analyzed. Model prediction results were evaluated with statistical

parameters such as root mean square error, mean absolute error, Akaike information criterion, Nash-

Sutcliffe efficiency, Kling-Gupta efficiency, r-square, average bias error, average bias error, bias factor.

In addition, the prediction results were compared with boxplot and timeseri plot in order to support the

statistical results. As a result of the analyses, the best predictions were obtained with the SFLA based

ANN model established with the Resilient Backpropagation activation function. In addition, the SFLA technique enabled the ANN model to produce the most accurate outputs by optimizing the parameters.

The outputs of the study help water resource managers and planners in estimating incomplete rainfall

and extending rainfall records.