Liegeman11/nigeria-avm

Nigeria's first open-source Automated Valuation Model. XGBoost trained on 1,635 scraped PropertyPro listings. MdAPE 28.4%. Live FastAPI REST API. # Nigeria Property AVM — Automated Valuation Model > ## Nigeria's first open-source Automated Valuation Model for residential property. > Built entirely from scraped data, deployed as a live REST API. ## The Problem Nigeria has no open land registry. No public transaction database. No Zillow. When a buyer, seller, bank, or investor needs to value a Nigerian property, they hire a human surveyor, a process that takes days and costs ₦50,000–₦200,000, with no transparency on how the number was reached. This project changes that. ## What I Built A full end-to-end machine learning pipeline that: 1. Scrapes live property listings from PropertyPro.ng (1,940 listings collected in V1; expanded to 4,143 in V2) 2. Cleans messy, inconsistent Nigerian property data (prices in mixed formats, unstandardised locations) 3. Engineers Nigerian-specific features — LGA-level price encoding, C of O status, BQ presence, premium area flags, LGA quality scores 4. Trains Ridge regression (baseline) and XGBoost (main model) 5. Deploys the model as a production REST API with confidence intervals and automatic documentation ## Live Demo ```bash curl -X POST localhost \ -H "Content-Type: application/json" \ -d '{ "bedrooms": 4, "bathrooms": 4, "lga": "Lekki", "state": "Lagos", "prop_type": "House", "has_cof_o": 1, "has_bq": 1, "has_pool": 0, "is_furnished": 0, "is_serviced": 0, "is_newly_built": 0, "is_gated": 0 }' ``` **Response:** ```json { "estimated_value_ngn": 391014048, "estimated_value_usd": 244383.78, "confidence_range_ngn": { "low": 312811238, "high": 469216857 }, "lga": "Lekki", "state": "Lagos" } ``` ## Model Performance (V1) | Metric | Ridge (Baseline) | XGBoost (Final) | Improvement | |---|---|---|---| | MAE | ₦212,938,822 | ₦195,281,484 | -8.3% | | RMSE | ₦417,689,904 | ₦398,752,440 | -4.5% | | MAPE | 59.5% | 50.9% | -8.6pp | | **MdAPE** | **34.1%** | **28.4%** | **-5.7pp** | > **Why MdAPE?** Nigerian property data contains extreme outliers. Med …