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.
