basilry/KB_Real_Estate_Historical_and_Predictional_Data

KB_Real_Estate_Historical_and_Predictional_Data

by. Kim Basilri

---

What is this?

• South Korean apartment records and predictions csv based on data from KB Real Estate Data Hub, and random natural language questions and parsing datasets based on them.
• Google Korea Organized Machine Learning Boot Camp After completion in 2024, data to be used for the project while participating in the working-level course of cooperation between the Ministry of Science and ICT of Korea, the Korea Information and Communication Industry Association, and Google Korea. It is being developed to expand and complete the existing Toadx2 project.

Configuration

1. kb_property_price_data.csv

   • Monthly and weekly sale price and jeonse price data for apartments in Korea from April 2008 to the present. In the case of monthly data, weekly data was converted and filled with a linear interpolation method.

2. kb_prdcition.csv

   • The sales price and jeonse price prediction data of Korean apartments about a year from this point on. Time series prediction was processed through Meta's Prophet model.

3. kb_region.csv

   • The region name and region code that are viewed in KB Real Estate Data Hub.

4. nlp_parsing_qna_dataset.json

   • Dataset to be used for learning and fine tuning of Gemma-2-2b-it models. The configuration is a dataset that randomly selects question templates based on kb_property_price_data.csv and kb_predicton.csv data, and parses the natural language in the desired form, assuming that the user asks the date and region and whether the sale/deposit is made.

Reference

1. KB Real Estate Data Hub's Apartment Dataset

2. ToadX2 Frontend Github

3. ToadX2 Backend Github

4. ToadX2 Web Page