Colombia Crop Yield Prediction

This project uses time series data to predict corn crop yield in Colombia.

Data

The training data was obtained from Datos Abiertos Colombia. All data related to corn crops were selected.

Methodology

• Data processing and visualization were performed with Pandas and Seaborn.

• Two-time series linear regression models were constructed using two data transformation processes:

  • Time-Step Feature
  • Lag Feature

Metrics

The models were evaluated using the following metrics:

• R-squared

• Mean Squared Error (MSE) and Root Mean Squared Error (RMSE)

• Mean Absolute Error (MAE)

• Residual Analysis

  • A visual analysis of the residuals was carried out using Matplotlib and Seaborn.

  • A quantitative analysis of the residuals was conducted using the Shapiro-Wilk normality test.

Conclusions

It was concluded that using either transformation technique for the model's input features (Time-Step Feature and Lag Feature) does not represent much variation in the model's results except for the p-value of the Shapiro-Wilk normality test which marked a slight difference.