The Toronto real estate price prediction

Actual vs Predicted Prices
Violin plot of price by building age
Violin plot of price per sift by building age
distribution of Property Prices
Residual Distribution
Feature importance
Partial dependence
Heatmap

Inspiration

By the high-demand real estate market of Toronto, we need a model to predict the price for properties which can help both buyers and sellers draw a rough picture of their potential transactions.

What it does

It was trained by multiple inputs and can predict the price of a property if given necessary factors.

How we built

We built it python. First we processed the data, deleting the useless rows and try to fill all the missing boxes in the column. And then we trained a xgboost, which is an optimized implementation of the gradient boosting framework designed for efficiency, flexibility, and portability. And we also added a module to demonstrate the prediction for new properties.

Output

We have trained a nice model whose r score is higher than 0.99(By using 80% of the data training the model and 20% of the data testing.) And we produced some really cool graphs to visualize the dependence of the price of different features. Along with the training model, we also came up with a heat map to visualize the output better.

Conclusion

The analysis provided a comprehensive understanding of the real estate market in Toronto, enabling the development of a robust solution for predicting property prices. By integrating geospatial data with traditional real estate data, the spatial dynamics of property values across the city were effectively captured. Methodologies such as spatial joins, iterative neighbor-fill, and feature engineering were both innovative and practical, ensuring the dataset was as complete as possible. This step was particularly important for maintaining the accuracy of the analysis, especially in areas with limited data. The code was well-structured, executable, and adhered to best practices, ensuring reproducibility and scalability. The visualizations and model evaluations generated during the analysis offered valuable insights into the data and the model's performance. These insights not only helped build an accurate predictive model but also provided actionable information for stakeholders in the real estate market. This project lays a strong foundation for future work in real estate price prediction and spatial analysis, making it a strong contender in the datathon.