Immo Eliza wanted a prediction tool to assist both buyers and sellers in Belgium to make market predictions of their real astate. Based on 6 features it will make the prediction
- area surface
- land surface
- number of bedrooms
- has open fire
- has swimming pool
- location
- Dataset: We used
data/properties.csvas datset. After the cleanup we held on to +/- 60k property listings in Belgium, houses & apartments. Thedata/cleaned_Dataset.csvis alrdy cleaned and has the best score.
- Tested Models: Linear Regression, Random Forest, XGBRegressor.
- Chosen Model: XGBRegressor was selected for its best score.
The XGBRegressor model achieved
- XGBoost Training Score: 0.9316877515763166
- XGBoost Test Score: 0.8129196050078894
- Mean Cross-Validation Score for XGBoost: 0.78153067585399
- Mean Absolute Error for test (MAE): 99075.4958276164
- Root Mean Squared Error for test (RMSE): 180832.77556397254
- Data Quality: Model accuracy relies heavily on data quality. Unique properties may be less accurately predicted.
- Time-Sensitive Factors: Does not consider market trends or economic conditions.
- Region Specific: Trained on Belgian data, which may not generalize to other regions.
- Explore more advanced models and feature engineering techniques.
- getting an Up-to-date dataset and clean it better.
Install dependencies from requirements.txt. Main libraries: pandas, scikit-learn, joblib, xgboost.
Run main.py to train the model. You can un-comment code if you want to experiment with better datacleaning. It will improve score, but it will also remove a lot of data.
Use predict.py. It will popup a form where you can fill in the correct features to get a prediciton in temminal screen