Used Mobile Phones Price Prediction for E-Commerce Application

Abstract

The second-hand mobile phone market is riddled with challenges such as inconsistent pricing and inaccurate condition evaluations, which impact buyer trust and seller profitability. This project aims to address these challenges by developing a comprehensive system for predicting the condition and price of used mobile phones, tailored to the Pakistani e-commerce market. We used a dual approach with the deep learning and the machine learning methodologies. During the condition prediction, the VGG16 model was selected as the best model with high accuracy across front on, front off and back view image classifications into Good, Average and Bad categories. An ensemble of machine learning models including XGBoost was used for price prediction; namely, XGBoost outperformed others by the R² score and error metrics, showing its strength in dealing with complex pricing features. Next, these models were integrated into a user friendly Flutter based mobile application, through which users could upload images and provide specifications for the real time predictions. It is an application that bridges usability and technology, and gives a scalable solution to market inefficiencies. Future work includes expanding the dataset to include additional phone models and market data, increasing the number of image views for condition assessment, and adapting the system for international markets. This project lays a strong foundation for creating transparency and consistency in the used mobile phone marketplace.