Multimodal price prediction

TL;DR

This paper proposes five deep learning models, including multimodal approaches, to predict cellphone prices based on specifications and images, demonstrating improved accuracy over existing methods with an 88.3% F1-score.

Contribution

It introduces novel multimodal deep learning models that combine graphical and non-graphical features for cellphone price prediction.

Findings

Multimodal models outperform unimodal approaches.

Achieved 88.3% F1-score in price prediction.

Using combined features improves prediction accuracy.

Abstract

Price prediction is one of the examples related to forecasting tasks and is a project based on data science. Price prediction analyzes data and predicts the cost of new products. The goal of this research is to achieve an arrangement to predict the price of a cellphone based on its specifications. So, five deep learning models are proposed to predict the price range of a cellphone, one unimodal and four multimodal approaches. The multimodal methods predict the prices based on the graphical and non-graphical features of cellphones that have an important effect on their valorizations. Also, to evaluate the efficiency of the proposed methods, a cellphone dataset has been gathered from GSMArena. The experimental results show 88.3% F1-score, which confirms that multimodal learning leads to more accurate predictions than state-of-the-art techniques.