Quantum256/Uganda-Quality-Coffee

The pipeline integrates Precision Agriculture Theory, DDQI, and GVC Theory to guide preprocessing, feature engineering, and modelling. It includes six phases: data profiling, theory-grounded imputation, construct-derived feature engineering, EDA, SMOTE handling, and ensemble modelling with SHAP & Uganda LOOCV. # Predicting Specialty Arabica Coffee Quality Using Ensemble Machine Learning ## Dataset **CQI Arabica Quality Database (May 2023)** Source: Kaggle – Coffee Quality Data CQI Sub-sample used: `df\\\_arabica\\\_clean.csv` — 207 records × 41 variables **Target Variable:** `Total Cup Points` (continuous regression) and `Specialty Grade` (binary: 1 if Total Cup Points ≥ 80, else 0) \--- ## Theoretical Frameworks This study is grounded in three complementary theoretical frameworks, applied analogously to how the UTAUT model was used by Kato et al. (2026) to organise and justify feature selection in MFA adoption prediction. ### 5.1 Precision Agriculture Theory Precision agriculture argues that within-lot variability in agronomic and post-harvest inputs is measurable, modelable, and actionable. Applied here, Total Cup Score is not random but is a predictable function of measurable inputs — altitude, processing method, moisture, and sensory attributes — that can be quantified and acted upon to guide quality upgrading. ### 5.2 Data-Driven Quality Intelligence (DDQI) Framework The DDQI framework posits that expert evaluator knowledge (Q Grader cupping scores encoded in the CQI database) can be extracted and deployed through machine learning, enabling quality prediction without physical expert presence. The ten CQI sensory dimensions represent structured domain knowledge that ensemble ML can encode and regularise. ### 5.3 Global Value Chain (GVC) Theory GVC theory (Gereffi et al., 2005) provides the economic rationale: specialty-grade certification is the mechanism by which producing-country actors capture value. A predictive model that identifies high-scoring lots and explains the drivers of quality translates into actionable intelligence for process upgrading (post-harvest) and product upgrading (consistent specialty access). \--- ## Construct Mapping Table Modelled after the UTAUT variable-to-construct mapping in Kato et al. (2026), the table below assigns eac …