This document describes a study that uses artificial neural networks (ANN) to predict moisture levels in the Chefchaouen region of Morocco. Specifically:
- ANNs with a multilayer perceptron (MLP) architecture were applied to predict moisture based on climatic variables like temperature, pressure, etc. collected over 1248 days.
- The optimal network structure was determined to have 4 neurons in a single hidden layer, with inputs normalized and outputs between 0-1.
- When trained on 70% of data, the ANN model achieved a correlation of 0.98 and mean squared error of 0.19%. On validation and test data, the correlation was 0.97 and error around 0.16