Crop Yield Prediction on an Agri Data Lake

A district-level crop yield predictor for India built on a three-zone data lake on a laptop CPU. ICRISAT district crop statistics, IMD gridded daily rainfall and temperature, soil properties and district boundaries are landed as delivered (bronze), joined by GIS into one row per district (silver) and turned into season features (gold); a LightGBM model then predicts the yield of rice, kharif sorghum, wheat and chickpea. On held-out years 2014-2017 it is 8.03 % off on average against 15.21 % for 'same as last year'. The work is delivered as AgriLake Yield, a secure web application with forecasts, what-if scenarios and maps.

Category: Machine Learning / Data Engineering / Geographic Information Systems / Agricultural Analytics

CSE Project 27 from ProTechSC Academy, with graphical abstract, video demonstration and project guidance.