Daniel-Abifarin/Financial-Inclusion-in-Africa-ML-project

Unsupervised ML project identifying financial excluded people among 23,524 East Africans by grouping them into 5 demographic segments # Financial-Inclusion-in-Africa-ML-project Unsupervised ML project identifying financial excluded people among 23,524 East Africans by grouping them into 5 demographic segments # Financial Inclusion in Africa — Demographic Clustering ## Overview Most machine learning projects on this dataset treat it as a classification problem such as predicting who has a bank account. I took a different approach. Instead of predicting a known outcome I wanted to let the data tell me who naturally groups together without giving the model any financial information at all. As a Nigerian, financial exclusion is not an abstract concept. The barriers I see across West Africa women locked out of formal banking, rural communities with no access, informal traders doing business without financial infrastructure are the same patterns I was looking for in this East African data. The question I wanted to answer was simple: who are the financially excluded and what do they actually have in common? To answer it I applied K-Means clustering with PCA dimensionality reduction to 23,524 individuals across Kenya, Rwanda, Tanzania and Uganda. I deliberately withheld the bank account column from the model, then checked it afterwards to see if the algorithm had discovered financially meaningful groups on its own. It had. ## Dataset - Source: Zindi Africa — Financial Inclusion in Africa Competition Original purpose: Supervised classification (predict bank account ownership) - My purpose: Unsupervised clustering (discover natural demographic segments) - Size: 23,524 individuals, 13 features - Countries: Kenya, Rwanda, Tanzania, Uganda - Data period:2016-2018 Finscope surveys ## Tools & Libraries - Python, Pandas, NumPy - Scikit-learn (KMeans, PCA, StandardScaler, silhouette_score) - Matplotlib, Seaborn ## Methodology ### 1. Exploratory Data Analysis - Examined distribution of all 13 features across 23,524 individuals - Confirmed no missing values - Only 14.1% of respondents have bank accounts.Thi …