Car price prediction (spb market of used cars)

What I did

1. First, I scraped data on cars from the website https://rolf-probeg.ru/spb/cars/ (pls, see auto_data.py file in src folder).
   Statistics for scrapped data before and after removing extreme values:

   Metrics	before	after
   count, q	2,518	2,384
   mean, RUB	2,099,148	1,862,896
   std, RUB	1,949,885	1,108,161
   min, RUB	97,200	400,000
   25%, RUB	1,039,000	1,049,000
   50%, RUB	1,598,283	1,576,749
   75%, RUB	2,469,947	2,350,138
   max, RUB	23,996,000	6,000,000

2. Second, I performed data cleaning, EDA and converted categorical data to numerical data (pls, see etl.py file in src folder. ETL was also performed in auto_rolf_june.ipynb file).

3. The next step was model selection. Initially I used LinearRegression, but it gave unsatisfactory result. So I tried RandomForestRegressor tuning with RandomizedSearchCV, GradientBoostingRegressor, LightGBM, CatBoostRegressor, XGboost.
   I chose LightGBM as this model is the fastest one and has pretty high accuracy. (pls, see auto_rolf.ipynb file)
   Pls, find below model's metrics:
   

   Metrics	Score
   r^2	0.892
   rmse	347 296.14 RUB
   mae	243 047.57 RUB

4. The final step was creating GUI with HTML and CSS.

You can see example in video below.

Car_price_prediction.mp4

Usage

• Clone this repo.
• Install the required packages (see requirements.txt).
• Run flask run or app.py.