A Recurrent Neural Network Model for Weather Forecasting


CEBECİ Y. E.

2019 4th International Conference on Computer Science and Engineering (UBMK), Samsun, Turkey, 11 - 15 September 2019 identifier identifier

  • Publication Type: Conference Paper / Full Text
  • Doi Number: 10.1109/ubmk.2019.8907196
  • City: Samsun, Turkey
  • Keywords: Data Mining, LSTM, Deep Learning, Adaptive Learning, RFE

Abstract

This paper compares data mining approaches for weather forecasting from one-dimensional and multidimensional meteorological weather data. Linear and nonlinear methods are applied and more successful results are obtained from nonlinear methods. The best result is obtained with LSTM(Long short-term memory). RFE(Recursive Feature Elimination) is used for subset feature selection and it increases one-dimensional MLP(Multi Layer Perceptron) model accuracy. In addition, Grid Search is used for hyperparameter tuning and early stopping is used to avoid overfitting and underfitting.