Logo Lanfrica
  • Accueil
  • Atlas
  • Analyses
  • Documentation
  • Sign in

© 2026 Lanfrica. Tous droits réservés. Tous les droits d'auteur des ressources affichées sur le site Web Lanfrica appartiennent aux détenteurs de droits d'auteur d'origine, sauf indication contraire explicite.

tanxsh/SCU-Analytics-Showdown-Spring-2026

Domaine:

agriculturesocioeconomic

Type de record:

model
Créateur:
tan
Hôte:
Winning submission for the SCU Analytics Showdown Spring 2026. Two-stage XGBoost hurdle model predicting farmer non-seller risk for Good Nature Agro across 22,597 farmers in Zambia. # **🏆 Winner — SCU Analytics Showdown, Spring 2026** ## Roots & Returns — The Story Every growing season, Good Nature Agro (GNA) extends agricultural input loans — seeds, fungicide, fertilizer, to thousands of smallholder farmers across Zambia. In return, farmers are expected to sell their harvest back to GNA at the end of the season. But **23.3% of farmers never sell back**. They take the inputs and disappear. This project answers the question GNA faces every season: **which farmers will we lose and what can we do about it before it's too late?** --- ## The Model Built a **two-stage XGBoost hurdle model** that separates the problem into two distinct questions: ``` Stage 1: Will this farmer sell back at all? → XGBoost Classifier (AUC: 0.880) Stage 2: If they do sell back, how much? → XGBoost Regressor (RMSE: 594 kg) Expected Yield = P(sells back) × E[yield | sells back] ``` This architecture handles the 23.3% zero-yield non-sellers that break a standard regression model. ### Risk Tiers Farmers are assigned to one of three risk tiers based on their expected yield: | Tier | Expected Yield | Actual Non-Seller Rate | Action | |------|---------------|----------------------|--------| | 🔴 **HIGH** | ≤ 83 kg | 54.5% | Immediate intervention | | 🟠 **MEDIUM** | 84 – 216 kg | 11.7% | Monitor and support | | 🟢 **LOW** | > 216 kg | 3.9% | Reliable | > By flagging just the top 27% of farmers by risk, the model catches **74% of all non-sellers**. --- ## Key Findings | # | Insight | Impact | |---|---------|--------| | 1 | Season 1 farmers have a **27.7% dropout rate** — 20× higher than Season 6+ farmers | Prioritise early-season outreach for first-timers | | 2 | **Fungicide** is the #1 controllable yield driver — Soy Bean farmers with fungicide yield 463 kg vs 160 kg without | Expand fungicide access; 69% of Soy Bean farmers don't receive it | | 3 | **Western and Southern regions** have 50%+ non-seller rates — drought-impacted crisis zones | Region-sp …

Visit

github.com

Similaires

jml0123/alp-spring-clientFentahun022/spring-ethiopian-validatorstchetagni/fintech-mobile-money-springAntonina-Sinitskaya/spring-otters-logistics-showcaseh3ymvnu/togo-geo-spring-boot-starterhamzaMissewi/Coldchain-Logistics-Full-Stack-Spring

jml0123/alp-spring-client

Client site for Kitabu (Microsoft / African Library Project hackathon) # Kitabu *(Client-site)* Kit

Fentahun022/spring-ethiopian-validators

# Ethiopian Validation Library for Spring Boot A collection of Ethiopia-specific validation annotat

tchetagni/fintech-mobile-money-spring

Spring Boot reference architecture for Mobile Money / Stripe / FlutterWave / Cyclos integration — Af

Antonina-Sinitskaya/spring-otters-logistics-showcase

Public portfolio case study of Spring Otters Logistics — a full-stack logistics marketplace for ship

h3ymvnu/togo-geo-spring-boot-starter

# Togo Geo — Spring Boot Starter Toute la **carte administrative officielle du Togo** (régions,

hamzaMissewi/Coldchain-Logistics-Full-Stack-Spring

A real-time cold chain logistics monitoring SaaS platform built specifically for Tunisian businesses