odoffin/financial-inclusion-analysis

Logistic regression analysis predicting bank account ownership in East Africa # Financial Inclusion in Africa — Logistic Regression Analysis ## Project Overview A machine learning analysis predicting whether individuals in East Africa have a bank account, based on demographic and socioeconomic survey data. Completed as part of the YMCA/Seneca Bridge to IT Careers programme (2026). Countries covered : Kenya, Rwanda, Tanzania, Uganda Survey years : 2016, 2017, 2018 Total respondents : 23,524 Target variable : bank_account (Yes/No) --- ## Business Question Can we predict whether a person has a bank account based on their age, education, job type, location and household characteristics? And which factors most strongly predict financial inclusion? --- ## Key Findings | Factor | Finding | |----------------------|--------------------------------------------------| | Class imbalance | 86% unbanked — addressed using SMOTE | | Best model | Random Forest with SMOTE oversampling | | Top predictor | Job type and education level | | Most banked group | Government employees (77.5% banked) | | Least banked group | No income group (2.1% banked) | | Gender gap | Males 19% banked vs females 11% | | Cellphone effect | Phone owners 10x more likely to be banked | | Best country | Kenya (25% banked) | | Worst country | Uganda (8.6% banked) | --- ## Models Compared Three models were trained and evaluated on the same data: - Logistic Regression (baseline — linear boundary) - Random Forest (200 trees voting — best performer) - Gradient Boosting (200 sequential trees — strong performer) All models trained on SMOTE-balanced data. All models evaluated on the real unmodified test set. --- ## Tools Used - Python 3 : programming language - pandas : data loading and …