Análisis de series de tiempo para predicción de variables climáticas mediante técnicas de aprendizaje automático y profundo

Abstract

Abstract: The present degree work aims to develop a climate prediction model using data provided by the Max Planck Institute for Biogeochemistry Beutenberg located in Jena - Germany, obtained in time series. It uses the KDD methodology to develop and evaluate two-period temperature and precipitation forecast models. ARIMA, LSTM, Simple Linear Regression and Multiple Linear Regression models are trained and evaluated using Python. The use of the sklearn library provided several useful functions with traditional machine learning algorithms. With the help of the open source framework TensorFlow it was possible to evaluate deep learning models. The evaluation of the results encompasses a comparison using the root mean square error metric RMSE and mean absolute percentage error MAPE. The results are then stored in a MySQL database. We chose to use the Django framework, as it integrates ideally with the aforementioned language. A dashboard created with Django visualizes the predictions and performance metrics, facilitating the analysis and interpretation of the data to improve the accuracy of weather forecasts.