NneamakaObiazor/gig_wallet_risk_analysis

A full analytics workflow built against a 50,000-transaction synthetic dataset covering gig-worker digital wallet activity across 4 African countries(Jan 2023–Dec 2024). # African Gig-Economy & Digital Wallet Risk — EDA, Verification & BI Model A full analytics workflow — Python EDA, statistical claim verification, a 5-page Power BI dashboard, and a polished analytical report — built against a 50,000-transaction synthetic dataset covering gig-worker digital wallet activity across Nigeria, Kenya, Ghana, and South Africa (Jan 2023–Dec 2024). The brief that shipped with this dataset proposed six specific fraud/risk patterns (channel effects, country concentration, new-account risk, gig segment disputes, velocity correlation, month-end spikes). **We tested all six directly against the data and none held up** — every categorical split and correlation came back statistically indistinguishable from noise (p > 0.05 in every case, most p > 0.4). Fraud, dispute, and reversal flags sit at a uniform ~50% no matter how you slice the data. This repo documents that verification process end-to-end, rather than writing a report that assumes the brief's claims are true, and carries the same verdict into the live Power BI file itself so the report and dashboard never disagree. See `report/Analytical_Report.pdf` for the full write-up, or `scripts/verification_tests.py` to reproduce the statistics yourself in under 10 seconds. ## Why this matters (the "AI-proofing" angle) A common failure mode when an LLM or analyst is handed a brief with pre-written "expected findings" is to produce a report that simply restates those findings with cosmetic numbers attached — regardless of what the data actually shows. This project is built to resist that: - Every claim in the original brief is run through an explicit statistical test (chi-square test of independence for categorical splits, Pearson correlation for the continuous velocity claim), with the test, statistic, p-value, and verdict shown side-by-side with the original claim. - The verification script (`scripts/verification_tests.py`) is a standalone, re-runnable harness — not just numbers pasted into a …