Forecasting of Corporate Revenues with Machine Learning Models versus Traditional Methods in the Digital Industry

Aug 4, 2026

Forecasting of corporate revenues is increasingly supported by machine learning models, with prior studies indicating potential advantages over traditional approaches. This thesis investigates three real-world datasets containing revenue and feature information for Amazon, Microsoft, and Netflix over the last two decades. The work applies seasonal differencing with logarithmic transformations, builds classical time-series baselines such as an autoregressive model of order 1, and trains machine learning alternatives including Partial Least Squares and deep neural networks. Model performance is assessed through metrics such as MAE and AIC, showing that AR(1) delivers the most accurate forecasting across all datasets, reducing error by more than 12% and up to 72% versus the best machine learning benchmarks in each dataset.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages58

Reading Time2 h 26 min

AccessFree to download

Introduction

Overview

Motivation

Research question and hypotheses

Literature review

Overview

Current trends in sales/revenues forecasting

On-demand ride services

E-commerce – fashion store

Retail industry – pharmacy retail sector

E-commerce – merchandise store

Publishing industry

Retail industry – retail store

Retail industry – Walmart

Retail food sector – bakery chain

Summary

Methodology, methods and tools

Methodology

Methods

Data collection and description

Data preprocessing

Cross Validation on time series

Time series models

Machine learning models

Forecasting of Corporate Revenues with Machine Learning Models versus Traditional Methods in the Digital Industry Joo Ribeiro dos Santos Pattenden 152021005 Dissertation written under the supervision of Professor Pedro Afonso Fernandes Dissertation submitted in partial fulfilment of requirements for the MSc in Business Analytics, at Catlica-Lisbon School of Business & Economics Forecasting of Corporate Revenues with Machine Learning Models versus Traditional Methods in the Digital Industry Joo Ribeiro dos Santos Pattenden Abstract There has been a growing interest in the applicability of machine learning models incorporate sales and revenues forecasting. Past research has found promising results in this field, which show that these models might outperform more traditional methods. In this thesis, three real-world datasets containing information about the revenues and other features of Amazon, Microsoft and Netflix in the last two decades are investigated to forecast the revenues of these digital companies. Firstly, we apply different pre-processing techniques on the data, which include seasonal differencing using logarithm transformations. Then, some more classical time-series methods including Autoregressive model or order 1 are built. Moreover, different machine learning models including Partial Least Squaresand Deep Neural network are applied. Finally, an empirical comparison of the models is performed using metrics such as Mean Absolute Error and Akaike Information Criterion. The results show that Autoregressive model of order 1 outperforms all the other models in terms of revenues forecasting accuracy in all datasets. Particularly, comparing with the benchmark machine learning model in each dataset, this method is able to reduce the error by more than 12 % and up to 72 % . Although these findings require further research to address any possible limitations, they provide insights on the performance of several models in revenues forecasting of digital firms, which can be a valuable tool for the decision-making process of businesses in this industry. Keywords: Machine Learning; Sales and revenues forecasting; Digital companies; Preprocessing; Time Series; Empirical comparison; Accuracy; Previs o de Receitas de Empresas com Modelos de Machine Learning versus Mtodos Tradicionais na Ind´ustria Digital Joo Ribeiro dos Santos Pattenden Resumo Tem havido um interesse crescente no uso de modelos de machine learning para a previso de vendas e receitas das empresas. Pesquisas recentes revelaram resultados promissores, que mostram que estes modelos podem superar mtodos de previso mais clssicos. Nesta tese, so investigadas trs bases de dados que contm informac¸ o sobre as receitase outros atributos das empresas Amazon, Microsoft e Netflix nas ´ultimas duas dcadas, com o objetivo de prever as suas receitas. Comec¸amos por aplicar diversas tcnicas de pr-processamento dos dados, que incluem as diferenc¸as homlogas de logaritmos. Posteriormente, so implementados alguns mtodos mais clssicos de previso de sries temporais como o modelo autorregressivo de ordem 1 . So desenvolvidos tambm diferentes modelos de machine learning como o modelo dos m´ınimos quadrados parciais e redes neuronais. Por fim, feita uma comparac¸ o dos modelos, utilizando diferentes mtricas como o erro mdio absoluto e o critrio de informac¸ o de Akaike. Os resultados mostram que o modelo autorregressivo de ordem 1 tem a melhor performance na previso das receitas nas trs bases de dados. Comparando com o melhor modelo de machine learning em cadauma das bases de dados, este mtodo consegue reduzir o erro em mais de 12 % e at 72 % . Embora estes resultados precisem de investigac¸ es adicionais, de modo a abordar poss´ıveis limitac¸ es, do-nos uma percec¸ o geral sobre o desempenho de diversos modelos na previso de receitas de empresas digitais, o que pode representar uma contribuic¸ o valiosa para a tomada de decises dos negcios desta i

FAQs

Which models are compared for forecasting corporate revenues in the thesis?

The thesis compares classical time-series methods such as an autoregressive model of order 1 with machine learning models including Partial Least Squares and deep neural networks.

What datasets are used to evaluate the forecasting performance?

Three real-world datasets are used, covering revenue and related features for Amazon, Microsoft, and Netflix across the last two decades.

What evaluation metrics are used to compare the forecasting models?

Performance is measured using metrics such as Mean Absolute Error (MAE) and Akaike Information Criterion (AIC).

Aug 4, 2026

Upload by Lucas Martin