Uso de redes neurais no cálculo da probabilidade de compra no cenário de comércio eletrônico brasileiro

Abstract

Nowadays, the evolution in data processing makes machines become much more agile and e cient when collecting, manipulating and generating information data. Associated with this, with the spread of the use of digital media such as the internet, smartphones and mobile devices, there is a great ease in the exchange of information between people around the world. In addition, essentially, any internet browsing generates usage traces, which are recorded by the servers of several companies interested in understanding more about the behavior of individuals. On the other hand, the e-commerce landscape is extremely challenging from a sales generation point of view. On average, one purchase is made for every one hundred visits registered on the main online sales platforms in Brazil. However, with so much information available to be used, within the context of Marketing 3.0, it is possible to associate the use of Neural Networks within the context of sales optimization. Using real data from a large sales platform, this work proposed to explore the scenario of the Calculation of Probability of Purchase within e-commerce, thinking about using this information to in uence the purchase journey of individuals and, using a Feed-forward neural network with one hidden layer, obtained results interesting both from a quantitative point of view, with 0.904 of accuracy, 0.903 of recall, 0.906 of precision and 0.905 of F1, and from a qualitative point of view, where the distributions of purchase and non-purchase probabilities were quite consistent with the reality of the events.