Developing front-end interface for model prediction here
The purpose of this project is to create a landing page data application for Bridgestone Realty, ltd. The model application, using proprietary data from the Paragon MLS portal, gives accurate listing evaluation of real estate properties within the Greater Vancouver Area to a Median Average Error of 50578 CAD, based upon custom imput from the users. The machine learning capabilities of the application and the consequent interface allows the model to constantly stay up-to-date on predictive real estate trend within Vancouver. Its ability to instantly provide accurate evaluation also creates a starting place for the realtor and any potential customers in further partnership.
- Bridgestone Property Corp, Ltd
- Partner contact: Liu Jian
- Inferential Statistics
- Machine Learning
- Data Visualization
- Predictive Modeling
- Sentiment Analysis
- Financial BERT
- Application deployment
- Sklearn
- Python
- Tensorflow, Pytorch
- PostGres, MySql
- Pandas, jupyter
- HTML
- Streamlit
- AWS EC2
Due to the unique nature of the Vancouver Real Estate Market, there is currently no commercial application of predictive property evaluation methods available for this segment . This lack of automated process for quick-and-easy evaluation contribute to the barrier many customers experience as they venture into the real estate market, where they have to right away make time-dependent commitment to specific agents without prior knowledge. From the perspective of the customer, there is also potential for conflict of interest when they approach any realtor for an evaluation of their desired property as a buyer, or evaluator of their exisitng property as a seller. The goal of this application, therefore, is to provide an expedient and objecive metric for those who are looking to make an important financial step in their lives. Just by visiting the landing page of Bridgestone Realty, Ltd, customers can have an accurate and objective metrics for their real estate need; this opens up the possibilities for their further interest or commitment with Bridgestone.
Using propietary database from the Greater Vancouver Area that have been properly anonymized to ensure security, the data is segmented and tested to ensure the integrity of the machine learning methods without any leakage. The data then undergoes a complex and comprehensive period of preprocessing that involves feature engineering and selection tools such as sentiment analysis, tokenization, and data aggregation.
Once completed, 22 machine learning algorithms and several iterations of different Feed Forward Neural Network architectures are compared to ensure maximum accuracy and relevance with the exisitng data. The winner of this process, a Random Forest ensemble method is chosen for further optimization and fine-tuning. The resultant product is then tested on previously unseen data where it reaches an average median error of 52230 CAD, which is within 3 percent of the average mean value of the testing target. An interface is build around the model for demonstration purposes, and an interface for predictive evaluation of properties is in its proof-of-concept stage.
- frontend developers
- data exploration/descriptive statistics
- data processing/cleaning
- statistical modeling
- writeup/reporting
- Model evaluation & comparison
- Data application deployment
- Clone this repo
- Raw Data is being kept here within this repo.
- Data processing/transformation scripts are being kept here
- Model generation scripts are being kept here
- For a demonstration of the interface, you can either view the static capture of the application here or contact me for a live demo.
There should be a minimal amount of setup required. Besides the standard libraries(Pandas, Sklearn, Etc.), all dependencies can be installed by just following the scripts alone
- Presentation on deliverables
- Keynotes and in-depth description of the app
- Documentation of model evaluation and testing
Team Leads (Contacts) : Jesse Lu(@Jesse Lu)