District-level hot/coldspot mapping for paddy, wheat & millets (2010–2019), based on Pramanik (2026), Agricultural and Resource Economics Review.
Paper's baseline is the 80th percentile; 70th/90th are its robustness checks.
All India — statistics computed nationally.
For the two t-test techniques, yield deviations, the t-test, and the percentile thresholds are all recomputed in your browser relative to the average of only your selected districts — a genuine regional analysis, not just a map filter. The Random Forest / Logistic Regression / Gradient Boosting / K-means models were trained once on the full national dataset (retraining per selection isn't done client-side); for those, your selection filters which districts are shown and summarized, using the nationally-trained predictions.
Random Forest, Logistic Regression and Gradient Boosting are trained to reproduce the statistical (percentile + t-test) label for each district, using features engineered from the yield series: Period 1 mean, Period 2 mean, change, % change, and within-period volatility. Accuracy shown is from stratified cross-validation. K-means instead clusters districts directly from these features into three yield tiers, with no reference labels.