Predicting Laptop Prices: An End-to-End Project Using E-commerce Website Data
In this project, I tried to create a model that can predict the price of a laptop based on the criteria of the desired laptop by using the data of the available laptops that I collected from the Digikala website, the largest e-commerce site in Iran.
Note: Since the price of laptops in Iran is constantly changing, this model is trained based on the data of laptops available on 2022-11-06 (1401-8-15) in Digikala. And in the coming years, it may not have an accurate prediction, and the solution to this problem is to update the model with the most up-to-date data.
Project Overview :
1 - Collect Data
To collect my data, I used Digikala's secret api. I was able to collect the data I wanted (available laptops with prices) with a simple fitler.
Collect Laptop main data
- id
- title_fa
- title_en
- price
- image_url
- brand
Collect Laptop details data
- cpu manufacturer
- cpu series
- cpu model
- ram
- ram type
- internal storage
- internal storage type
- gpu manufacturer
- gpu model
- screen resolution
- ports
Merge Collected data
Remove duplicated rows
Save data into csv file
2 - Take a look at data :
After collecting the data, I started checking the collected data to make sure it was collected correctly
- #### Check shape of data
- #### Check is there any null value
- #### Check data types
- #### Check number of unique values in each column
3 - Cleaning data
Like all machine learning projects, the data doesn't arrive perfect and ready for prediction. At this point, I started cleaning the collected data.
- #### Convert brands name from persian to english
- #### Convert ram from persian to english digits
- #### Clean and convert internal storage to english
- #### Convert and clean internal storage to english
- #### Convert and clean laptops screen size
- #### Clean laptops resolution
4 - EDA
For the next step, which is Feature engineering, it was necessary to get information about the data. In this step, I analyzed and explored the data.
- #### Laptops price distribution
- #### Number of laptops of each brand
- #### Number of cpu of each cpu manufacturer
- #### Number of laptops for each ram group
- #### Number of laptops for each ram type group
- #### Number of laptops with diffrent internal storage
- #### Number of laptops for each internal storage group
- #### Number of laptops with diffrent screen sizes
- #### Number of laptop with diffrent screen resolution
5 - Feature Engineering
In this step, I prepared the features for training the model
- #### Remove outliers base on laptops price using z score
- #### Convert screen resolution to number
- #### Extract Gaming brands from title (asus rog , acer nitro ...)
- #### Remove brand with only one laptop
- #### Extract clean gpu model from gpu model column
- #### Remove laptops with only less than 3 model gpu
- #### Label endcoding cleand gpu models
- #### Convert internal storage from tb and gb to mg
- #### Label encoding internal storage type
- #### Convert ram from str to int
- #### Extract port count
- #### Label encoding ram type
- #### Label encoding cpu series
- #### One hot encoding brand - cpu manufacturer - gpu manufacturer (nominal categorical variables)
6 - Feature Selection
In this step, I chose the features needed to train the model
- #### Check correlation
- #### Mutual information regression
7 - Model training
8 - Hyperparameter tuning
In this step, I hyperparameter tuned the model that had the best r2 score
9 - Cross validation
10 - Save model
I pickeld model for use in the gui environment
11 - Website
To create website, I used streamlit formwork, a powerful formwork that allows me to create the desired user interface completely using Python.
12 - Deploy
I used Heroku a cloud platform as a service which provide a free hosting to deploy my app on it. it's and amzaing platform gave me so much flexbilte to deploy your apps
