Image-Driven Used iPhone Damage Assesment and Price Prediction with Explainable AI

Abstract

The used smartphone market, particularly for iPhones, faces significant challenges related to fair price estimation due to various influencing factors such as physical damage and market trends. Current valuation processes lack transparency and often result in iPhone user dissatisfaction, benefiting resellers at the expense of iPhone users. Additionally, deep learning models for price prediction and damage assessment are usually ""black box"" in nature, making their decision-making processes challenging to interpret. The research proposes a comprehensive approach combining deep learning models and machine learning models with explainable AI (XAI) for damage detection and price prediction of used iPhones. The project includes the development of a Convolutional Neural Network (CNN) for mobile screen damage assessment and a comparative study among machine learning models , shallow neural networks like General Regression Neural Network (GRNN) and deep learning models like MLP regressor for predicting iPhone prices based on conditions, specifications and market trends. XAI techniques such as SHAP and Grad-CAM are proposed to integrate to enhance the model’s interpretability, providing visual explanations for damage detection and feature contributions in the price prediction model. The final evaluation of the author's research demonstrated a robust performance of damage detection and price prediction models. The damage detection model developed by the author for damage detection achieved a validation accuracy of 97%. In evaluating models for price prediction, the author identified the Multi-Layer Perceptron (MLP) regression model as the most accurate, achieving an R² value of 0.9913. The linear regression exhibited minimal prediction errors, highlighting its robust predictive capability.