charisquaye/Graph-Based-Supply-Chain-Optimization-for-Cocoa-Farmers-in-Ghana

# Graph-Based Supply Chain Optimization for Cocoa Farmers in Ghana Reproducible computational pipeline accompanying the thesis of **Emmanuel Mensah Quaye** (research methodology and results chapters). ## Overview This repository contains the complete analysis pipeline that: 1. Generates a synthetic, calibrated representation of Ghana’s cocoa supply chain (900 farmers, 45 buying points, 10 depots, 2 ports). 2. Builds a weighted, multi-relational directed graph. 3. Applies centrality analysis, Dijkstra least-cost routing, and Louvain community detection (resolution = 1.0). 4. Trains Random Forest and XGBoost models for farm-gate price with and without network-derived features (5-fold CV, held-out test evaluation). 5. Runs a 5,000-iteration Monte Carlo simulation of baseline vs. consolidated marketing configurations. 6. Supports a prototype decision-support interface (Cocoa Network). All stochastic components are governed by the fixed master seed **20250115**. Re-running the pipeline regenerates the numerical results reported in Chapter 4 exactly. ## Quick start ```bash python -m venv .venv source .venv/bin/activate # or .venv\Scripts\activate on Windows pip install -r requirements.txt python analysis_chapter4_pipeline.py ``` Outputs: - `results.json` – all numeric summaries - intermediate pickles / npy files used for figures ## Key parameters (see also Table 4.1 of the thesis) | Parameter | Value | |-----------|-------| | Master seed | 20250115 | | Farmers / buying / depots / ports | 900 / 45 / 10 / 2 | | Edge-weight coefficients (α, β, γ, δ) | 1.0, 0.8, 0.6, 25.0 | | Louvain resolution | 1.0 | | Cross-validation folds | 5 | | Monte Carlo iterations | 5 000 | | Central consolidation saving / premium | 35 % / 0.30 GHS kg⁻¹ | ## Prototype decision-support tool (live) **Cocoa Network** – interactive decision-support prototype aligned with Chapters 3–4: - **Live URL:** cocoa-network-decision-supp… Sections include evidence overview, …