Regression model created to predict the price of the used phones. Deployed in streamlit cloud. Created using Kaggle noteboook
Dataset: https://www.kaggle.com/datasets/ahsan81/used-handheld-device-data
Kaggle link: https://www.kaggle.com/kavya2099/used-phone-price-prediction
Streamlit link: https://used-phone-price-prediction.streamlit.app/
Used phones and tablets have become increasingly popular in both the global and Indian markets in recent years. The rise of e-commerce platforms, improved quality of refurbished devices, and the increasing affordability of technology have all contributed to this trend.
Around the world, consumers are realizing the many benefits of purchasing used devices. Not only do these devices offer a more affordable alternative to brand-new gadgets, but they also have a smaller environmental impact. Many consumers are also attracted to the idea of reducing waste by giving new life to a previously-owned device.
In addition to being more affordable, used phones and tablets in India often come with a warranty, providing peace of mind to consumers who may be worried about purchasing a previously-owned device. This, combined with the improved quality of refurbished devices, has made used phones and tablets a popular choice for many consumers in India.
In addition, companies that collect and refurbish used devices can also collect valuable data and insights about consumer behavior and preferences, which can inform their future product development and marketing efforts. Overall, the collection and reselling of used phones and tablets can be a profitable and socially responsible business model for companies.
As a data scientist working at a company specializing in the collection and resale of used phones and tablets, the goal is to build a regression model to predict the used price of these devices. This information is crucial for the company to effectively price their devices and maximize their profits.
The dataset contains information about the used phone prices and tablets, including the model, OS, battery, screen size, storage capacity, and other relevant features. The task is to perform exploratory data analysis (EDA) on the dataset to understand the underlying relationships and patterns in the data, and then build a regression model to predict the used price of the devices.
The final model should be highly accurate and able to generalize well to unseen data. This project will provide valuable insights into the used phone and tablet market and enable the company to make informed pricing decisions.
• What is the problem you're trying to solve? The goal is to build a regression model to predict the used price of these devices.
• What data do you have available? Data containing the information about the used phone prices and tablets, including the model, OS, battery, screen size, storage capacity, and other relevant features.
• What are the potential solution(s) you can implement? Analyzing the features to find out the pattern and relationhip with the target variables that could help in getting better evaluation results
• What performance metrics will you use to evaluate your solution? Using MSE and R2 score to evaluate the regression model
• What techniques will you use?
* Data processing
* EDA
* Feature Engineering
* Scaling and Normalization
* Model Selection and Evaluation
* Test the model
• How will you validate your solution? Model which provides low MSE and high R2 score close to 1 would be validation point to select the best model
- Samsung was the most reused phone next to phone brands which are categorized as Others.
- There are more people using Android OS
- Screen size is mostly between 10-15 cm
- 4g and 5g users are more
- Rear camera mp range is mostly between 5-15 mega pixels
- Front camera mp range is mostly between 0-10 mega pixels
- Internal memory is mostly between 0-100 GB
- Phones ram storage is mostly between 3-5 GB
- Battery life is mostly around 3000 mAh
- Weight is around 150-210 grams
- Most of the phones were released between 2013-2015
- People have used the phones on an average between 600-800 days which is approximately 2-2.5 years
- Average normalized used price was between 4.2-5 k
- Average normalized new price was between 5-6 k
- Most of the features has rising relationship with the target variable --> normalized used price
- Linear Regression with top 5 features was considered to be the best model for this problem.
- Implemented the model in Streamlit.
- Given that the dataset indicated data collection from Asian and European markets, assuming the normalized prices were in Euros, we converted them to enable the model to predict values in Indian rupees.
- While the target feature is the normalized used price, we transformed the values back to actual prices for user display purposes.
- Plotting histogram using subplots makes visualization easier
- If we need to remove the last plot in a subplot for the visualizations which has odd number(eg: 11 plots but subplots shows 12), we can remove last one using delaxes
- Using regplot on bivariate analysis helps us to identify the rising/falling relationship with target variables
- SelectKBest in feature selection can be used to pick up top K number of features which can used in model training
- Applying crossvalidation concepts in model training
- Applying Pipeline concepts in model training
- Using Q-plots to visualize the distribution of a dataset and to compare it to a theoretical distribution
- Trying out other Feature selection methods
- Using Z scores to treat outliers