Machine Learning and Deep Learning for Rainfall Prediction
Sep 2023 – Mar 2024 · École polytechnique, France · Supervisor: Prof. Julien Chiquet
- Time series
- LSTM / CNN
- Interpretability
Research project of the École polytechnique curriculum (EA initiation à la recherche), with Léa Bohbot and Kenta Vert, supervised by Prof. Julien Chiquet (Université Paris-Saclay, INRAE, AgroParisTech).
We adapted an LSTM network and a CNN to temporal sequences of daily weather observations to predict rainfall. After a theoretical and practical study of LSTMs, we compared their results with standard machine-learning methods (logistic regression, random forests, XGBoost…). We then turned to the interpretability and simplification of the model — identifying the most explanatory variables — an important step to ensure the reliability of the predictions.
Code: LSTM · CNN · Report (in French): PDF
