0
- Volume 3, Issue 2 2025
By Syed Farasat Ali
DOI:10.20547/aibd.253204
Keywords: Keywords: e-Commerce, Sales Forecasting, LSTM, GRU, Session Level Forecasting
Abstract: This paper talks about a novel comparison of two deep learning models namely Long short-term memory (LSTM) and Gated recurrent unit (GRU). The target domain is e-commerce sales prediction applied on behavioral data from a Shopify store. The dataset comprises 87000 records stepped by hours. The dataset contains details about various features such as session duration, bounce rate, cart addition, checkout activity. It has been found that LSTM and GRU both were able to learn the spatiotemporal dependencies present in the data. These models predict with good accuracy the probability of making a purchase in a session. By means of extensive training, it has been found that both the models achieve good accuracy, precision, recall and F-1. Nonetheless, the LSTM model was able to achieve slightly better performance in terms of the above metrics. It can be concluded that the sequence-to-sequence models can be used for decision making in the domain e-commerce sales. In addition, strategic planning can be performed using AI models. The findings of the study can be used as evidence for scalable deployment of AU models in businesses so that they can respond well to consumer behavior.
Submission Date: 13 Aug, 2025 Reviews Completed: 22 Nov, 2025Acceptance Date: 26 Nov, 2025 Publication Date: 31 Dec, 2025