A Mobile-Based Deep Learning Model for Product Classification in Support of Boycott Campaigns

The boycott has had a significant impact on the country's global economy. Due to recent events in Palestine, support for the Palestinians was channeled through a boycott. This made it difficult for many people, particularly those with poor vision, to determine whether a product had been boycotted. Previously, such people would ask others in supermarkets, but this method proved ineffective because not everyone was aware of boycotted products. This paper proposes a new mobile-based deep learning model for product classification to benefit Palestinian causes and Egyptian trade. It consists of two main parts; the front-end part and the back-end part. In the front-end part, the user takes an image of a product using the mobile device. Then, this image is classified as either boycotted or not boycotted. If a product is boycotted, alternative products are suggested. The back-end part consists of three phases: data preprocessing, classification, and evaluation. In the preprocessing phase, data augmentation techniques and data oversampling methods are applied. Then, the processed data is used to feed the utilized deep learning architectures. Finally, several evaluation measurements are adopted to evaluate the performance of the proposed model. Two datasets are used: a private dataset containing both boycotted and non-boycotted products, and a benchmark dataset that includes a variety of grocery categories. The proposed mobile-based deep learning model achieves 94% accuracy on the private dataset, with an F1 score of 85%, precision of 92%, sensitivity of 80%, specificity of 98%, and area under the curve (AUC) of 98%. On the bench market dataset, the proposed model achieves 87% accuracy, with an F1 score of 87%, precision of 87%, sensitivity of 87%, specificity of 100%, and an AUC of 98%. The experimental results showed that Xception is the best deep learning architecture for the private dataset, and InceptionV3 is the best deep learning architecture on the benchmark dataset.