VAL-Jerono/KHS_housing_dissertation

This dissertation models household-level financial vulnerability and housing risk across all 47 Kenyan counties using the 2023/24 KNBS Kenya Housing Survey microdata. # Predicting Housing Financial Vulnerability Across Kenya Using Machine Learning on the 2023/24 Kenya Housing Survey **MSc Dissertation — Data Science & Analytics** **Strathmore University, Nairobi, Kenya** **Student:** VAL Jerono **Repository:** github.com **Deployment:** *(Streamlit / Hugging Face — link to be added upon launch)* --- > *"You cannot protect what you cannot measure."* > This dissertation argues that housing financial vulnerability — the silent precursor to eviction, uninsured loss, and generational poverty — can be systematically measured, mapped, and predicted from nationally representative survey data, enabling evidence-based resource allocation across Kenya's 47 counties.* --- ## Table of Contents 1. Abstract 2. Introduction 3. Literature Review 4. Methodology - 4.1 Data Collection and Understanding - 4.2 Data Preparation - 4.3 Exploratory Data Analysis - 4.4 Machine Learning Modelling - 4.5 Performance Evaluation - 4.6 Optimisation - 4.7 Deployment 5. Results 6. Discussion 7. Repository Structure 8. How to Reproduce 9. Applications and Stakeholders 10. References --- ## 1. Abstract ### Background and Research Problem Kenya's housing sector is marked by a structural paradox: approximately 61% of urban households live in informal settlements, yet housing financial vulnerability — the compounded risk arising from rent burden, insecure tenure, poor dwelling quality, environmental hazard exposure, and utility deprivation — remains unmeasured at a granular, data-driven level. Existing instruments such as insurance loss ratios and poverty headcounts capture isolated dimensions but offer no integrated, actionable risk score that practitioners, policymakers, and actuaries can operationalise at household or county scale. ### Method This study leverages the 2023/24 Kenya Housing Survey (KHS), a nationally representative microdata dataset comprising 21,347 households across all 47 counties, collected between 2 …