Modelo de predicción explicable basado en redes neuronales para el dominio de e-commerce

Abstract

Abstract: E-commerce platforms produce a large amount of data that can be used to make predictions about sales, product demand and customer behavior. However, models based on neural networks, despite having high performance, face criticism due to their black box nature, making it difficult to interpret the decisions made. This work aims to integrate explainability techniques, such as SHAP, to interpret, but also justify the predictions taken from these models in the e-commerce domain. A public e-commerce dataset containing product, user and transaction data was selected using the CRISP-DM methodology. Through exploratory analysis and the implementation of a neural network model, personalized product predictions were generated. After this, explainability techniques were used to decompose and analyze the impact of the variables on the model's decision making, achieving trust and transparency in their use.