Name: | Description: | Size: | Format: | |
---|---|---|---|---|
1.58 MB | Adobe PDF |
Authors
Advisor(s)
Abstract(s)
A gestão eficiente das contas a receber é essencial para a saúde e estabilidade financeira de uma empresa, uma vez que afeta diretamente a liquidez, a estabilidade operacional e a capacidade de cumprir obrigações futuras. Esta dissertação recorre à utilização de machine learning para efetuar previsões de contas a receber, destacando a importância dos modelos de regressão na previsão dos resultados do fluxo de caixa. Assim, o objeto de estudo deste projeto é a análise de vários algoritmos de regressão, como por exemplo K-Nearest Neighbors, Gradient Boosting Machines, Extra Trees, Random Forest, XGBoost, Árvores de decisão,
Regressão linear, Lasso e Ridge, e a escolha do algoritmo com melhores métricas. O objetivo final do projeto é prever o número de dias que o cliente demora a efetuar a liquidação dos documentos desde a data de vencimento dos mesmos de forma a melhorar a visibilidade do cash flow operacional. O modelo criado neste estudo ajuda a ter um maior controlo sobre o fluxo de caixa operacional e uma gestão mais eficaz das contas a receber, fazendo com vista a melhorar a tomada de decisão e maximizar a eficácia do planeamento financeiro. O estudo utiliza a metodologia CRISP-DM, que passa por várias fases como o pré tratamento de dados, feature engineering e a otimização de modelos, assegurando o alinhamento entre os procedimentos técnicos e o objetivo final. De forma, a melhorar os hiperparâmetros do modelo e, consequentemente atingir um melhor desempenho na previsão do conjunto de dados, foi aplicada a otimização bayesiana. O modelo desenvolvido utiliza dados reais de uma empresa produtora integrada de floresta, pasta, papel, tissue, soluções sustentáveis de packaging e bioenergia, para efetuar as
previsões do número de dias que um cliente demora a liquidar os seus documentos desde a data de vencimento, tendo em conta, o tipo de negócio, a divisa e o país do mesmo.
Efficient management of accounts receivable is essential for a company's financial health and stability, as it directly affects liquidity, operational stability and the ability to fulfil future obligations. This dissertation uses machine learning to forecast accounts receivable, emphasising the importance of regression models in predicting cash flow results. Thus, the object of study of this project is the analysis of various regression algorithms, such as K-Nearest Neighbours, Gradient Boosting Machines, Extra Trees, Random Forest, XGBoost, Decision Trees, Linear Regression, Lasso and Ridge, and the choice of the algorithm with the best metrics. The ultimate goal is to predict the number of days it will take the customer to settle the documents by the due date in order to improve the visibility of the operational cash flow. The model created in this study helps to gain greater control over operational cash flow and more effective management of accounts receivable, improving decision-making and maximising the effectiveness of financial planning. The study uses the CRISP-DM methodology, which goes through several phases such as data pre-processing, feature engineering and model optimisation, ensuring alignment between the technical procedures and the final objective. In order to improve the model's hyperparameters, Bayesian optimisation was applied to achieve better performance in predicting the data set. The model developed uses real data from an integrated forestry, pulp, paper, tissue, sustainable packaging solutions and bioenergy company to forecast the number of days it takes for a customer to settle their documents by the due date, taking into account the type of business, currency and country.
Efficient management of accounts receivable is essential for a company's financial health and stability, as it directly affects liquidity, operational stability and the ability to fulfil future obligations. This dissertation uses machine learning to forecast accounts receivable, emphasising the importance of regression models in predicting cash flow results. Thus, the object of study of this project is the analysis of various regression algorithms, such as K-Nearest Neighbours, Gradient Boosting Machines, Extra Trees, Random Forest, XGBoost, Decision Trees, Linear Regression, Lasso and Ridge, and the choice of the algorithm with the best metrics. The ultimate goal is to predict the number of days it will take the customer to settle the documents by the due date in order to improve the visibility of the operational cash flow. The model created in this study helps to gain greater control over operational cash flow and more effective management of accounts receivable, improving decision-making and maximising the effectiveness of financial planning. The study uses the CRISP-DM methodology, which goes through several phases such as data pre-processing, feature engineering and model optimisation, ensuring alignment between the technical procedures and the final objective. In order to improve the model's hyperparameters, Bayesian optimisation was applied to achieve better performance in predicting the data set. The model developed uses real data from an integrated forestry, pulp, paper, tissue, sustainable packaging solutions and bioenergy company to forecast the number of days it takes for a customer to settle their documents by the due date, taking into account the type of business, currency and country.
Description
Keywords
Fluxo de caixa Contas a receber Aprendizagem automatizada Previsão Cash flow Accounts receivable Machine learning Forecasting