Luyipai Real Estate Valuation System

Project Background:

• Automated property valuation for assets on Luyipai platform.
• Addressed inefficiencies and low accuracy in manual property appraisals.

Data Collection & Processing:

• Data sources: Transaction data, listing data, neighborhood prices.
• Data cleaning: Outlier removal (3-sigma rule), missing value imputation, log transformation for normality.
• Feature engineering: Area segmentation, building age, decoration, floor, etc.

Modeling & Optimization:

• Regression algorithms: Linear Regression, Random Forest, GBDT, XGBoost, LightGBM.
• Hyperparameter tuning: Grid search + cross-validation for tree depth, number of trees, learning rate, min samples per leaf.
• Ensemble: Stacking of GBDT/XGBoost/LightGBM with linear regression.
• Addressed imbalance with resampling and probability smoothing.

Workflow:

1. Process raw data and feature engineering.
2. Train multiple regression models with optimized hyperparameters.
3. Fuse results from listing and transaction data.
4. Detect and correct valuation anomalies.

Project Outcome:

• Decision tree regression MAPE: 92%; stacking improved by 3%.
• Increased efficiency and reliability of property valuation, aiding auction asset assessment.