O uso de machine learning na previsão de vendas

Abstract

Recently, company ABC, which operates in the B2B market segment and offers digital transformation services, has seen an increase in the volume of Deals and Leads. This increase in Deals and Leads has resulted in an increase in the revenue of company ABC. In order to predict the revenue of company ABC, a quantitative approach was used, where the efficiency of the machine learning algorithms Multiple Linear Regression (MLR), Support Vector Regression (SVR) and Multilayer Perceptron (MLP) in predicting the monthly revenue was compared, using data from the company between 2017 and the beginning of 2023. The comparison of the algorithms efficiency was based on two techniques: (i) the arbitrary division of the database into training base and test base and (ii) the K-Fold cross validation technique. Using the first technique, the MLP model was found to be the most suitable to predict the monthly revenue of company ABC, considering the performance measures R 2 and Mean Square Error (MSE). However, using the K-Fold cross validation technique it is concluded that the SVR model performs better in predicting the monthly revenue of company ABC. Considering that the K-Fold cross validation technique uses the entire database to train and test the algorithms, and in a way, mitigates the occurrence of overfitting, it may be fairer to compare the models based on this technique. Therefore, the SVR model seems to be the most suitable algorithm to predict the monthly revenue of the company ABC.

Date of Award	20 Sept 2023
Original language	Portuguese
Awarding Institution	Universidade Católica Portuguesa
Supervisor	Joana Pinho (Supervisor) & Miriam Taís Salomão (Co-Supervisor)