Reconstructing incomplete agricultural production series

Official agricultural statistics for Cameroon’s Centre region are incomplete, and the gaps are not random. Training a model on what remains bakes the collection failures into the predictions.

This project builds an ensemble of statistical reconstruction methods — spline interpolation, regression, k-nearest neighbours and ratio-based estimation — combined under hard agronomic constraints so that no reconstructed value can exceed what is physically plausible for the crop and area. The result fills interior gaps and extends the series retroactively for maize and onion.

Reconstructed values are flagged as such throughout, so downstream analysis can weight or exclude them.