| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 4.87 MB | Adobe PDF |
Autores
Orientador(es)
Resumo(s)
Este estudo investiga a aplicação de técnicas de Machine Learning na previsão de rotas, com foco
no comportamento do motorista, utilizando uma base de dados fornecida por uma empresa de
logística. O comportamento dos motoristas é influenciado por vários fatores, como preferências
pessoais, experiência com determinadas rotas, condições do tráfego e características das entregas,
como a distância, o peso da mercadoria, a urgência e a frequência de visita a um cliente específico.
Estas variáveis tornam o planeamento de rotas uma tarefa complexa, mas essencial para a
eficiência logística.
O principal objetivo deste trabalho foi desenvolver um sistema preditivo robusto e eficiente que
pudesse não apenas prever as rotas mais prováveis com base nas escolhas históricas dos
motoristas, mas também otimizar essas rotas com base em múltiplos atributos operacionais. Estes
atributos incluem a distância total percorrida, o peso da carga, a capacidade do veículo, e em alguns
modelos a frequência de visita ao cliente. A adição da frequência de visita a certos clientes foi
particularmente importante para aumentar a precisão das previsões. Esses atributos influenciam
diretamente a escolha de uma rota, uma vez que os motoristas tendem a preferir rotas mais
conhecidas para clientes frequentes ou ajustar as suas escolhas com base no peso e no volume das
entregas, visando minimizar o esforço físico e o tempo de transporte.
O estudo iniciou-se com a organização dos dados em conjuntos de treino e teste, permitindo a
aplicação e a validação dos modelos de Machine Learning. Diversas técnicas foram exploradas,
incluindo modelo de Osquare juntamente com técnicas de Regressão Linear, Random Forest, Redes
Neurais e Support Vector Machine (SVM). A eficácia desses modelos foi avaliada através de
métricas como Kendall's Tau, Accuracy e Edit Distance, permitindo uma análise comparativa dos
resultados.
Os resultados indicaram que os modelos de regressão com Redes Neurais, com a adição da
frequência como um atributo, se destacaram em termos de precisão e eficiência geral, obtendo
uma acurácia média de 0,9603, e um Kendall médio de 0,0314. O modelo Random Forest também
apresentou um bom desempenho, particularmente quando otimizado e com o atributo adicional
da frequência de visita, atingindo uma acurácia média de 0,9583 e um Kendall médio de 0,333. Em
contrapartida, técnicas como a regressão logística e SVM mostraram-se menos eficazes em certos
cenários.
Este estudo demonstra o potencial do Machine Learning para otimizar processos logísticos,
destacando a importância de uma análise detalhada e criteriosa dos dados. As descobertas
oferecem contribuições significativas para o campo da logística, mostrando como a integração de
técnicas avançadas de aprendizagem automática pode ser aplicada de forma eficaz no contexto
industrial, ao mesmo tempo em que aponta caminhos para melhorias futuras.
This study investigates the application of Machine Learning techniques in route prediction, with a focus on driver behavior, using a dataset provided by a logistics company. Driver behavior is influenced by various factors, such as personal preferences, experience with specific routes, traffic conditions, and delivery characteristics, including distance, cargo weight, urgency, and the frequency of visits to a specific customer. These variables make route planning a complex but essential task for logistical efficiency. The main objective of this work was to develop a robust and efficient predictive system that could not only forecast the most likely routes based on drivers’ historical choices but also optimize these routes based on multiple operational attributes. These attributes include the total distance traveled, load weight, vehicle capacity, and, in some models, customer visit frequency. The addition of visit frequency to certain customers was particularly important for improving prediction accuracy. These attributes directly influence route choices, as drivers tend to prefer familiar routes to frequent customers or adjust their choices based on the weight and volume of deliveries to minimize physical effort and transport time. The study began by organizing the data into training and testing sets, allowing for the application and validation of the Machine Learning models. Various techniques were explored, including Osquare modeling along with Linear Regression, Random Forest, Neural Networks, and Support Vector Machine (SVM) techniques. The effectiveness of these models was evaluated through metrics such as Kendall’s Tau, Accuracy, and Edit Distance, allowing for a comparative analysis of the results. The results indicated that regression models with Neural Networks, with the addition of frequency as an attribute, stood out in terms of overall precision and efficiency, achieving an accuracy of 0.9603 and a Kendall of 0.0314. The Random Forest model also performed well, particularly when optimized and with the additional frequency attribute, reaching an accuracy of 0.9583 and a Kendall of 0.333. In contrast, techniques such as logistic regression and SVM were less effective in certain scenarios. This study demonstrates the potential of Machine Learning to optimize logistical processes, highlighting the importance of a detailed and careful data analysis. The findings offer significant contributions to the field of logistics, showing how the integration of advanced Machine Learning techniques can be effectively applied in an industrial context while also pointing to avenues for future improvements.
This study investigates the application of Machine Learning techniques in route prediction, with a focus on driver behavior, using a dataset provided by a logistics company. Driver behavior is influenced by various factors, such as personal preferences, experience with specific routes, traffic conditions, and delivery characteristics, including distance, cargo weight, urgency, and the frequency of visits to a specific customer. These variables make route planning a complex but essential task for logistical efficiency. The main objective of this work was to develop a robust and efficient predictive system that could not only forecast the most likely routes based on drivers’ historical choices but also optimize these routes based on multiple operational attributes. These attributes include the total distance traveled, load weight, vehicle capacity, and, in some models, customer visit frequency. The addition of visit frequency to certain customers was particularly important for improving prediction accuracy. These attributes directly influence route choices, as drivers tend to prefer familiar routes to frequent customers or adjust their choices based on the weight and volume of deliveries to minimize physical effort and transport time. The study began by organizing the data into training and testing sets, allowing for the application and validation of the Machine Learning models. Various techniques were explored, including Osquare modeling along with Linear Regression, Random Forest, Neural Networks, and Support Vector Machine (SVM) techniques. The effectiveness of these models was evaluated through metrics such as Kendall’s Tau, Accuracy, and Edit Distance, allowing for a comparative analysis of the results. The results indicated that regression models with Neural Networks, with the addition of frequency as an attribute, stood out in terms of overall precision and efficiency, achieving an accuracy of 0.9603 and a Kendall of 0.0314. The Random Forest model also performed well, particularly when optimized and with the additional frequency attribute, reaching an accuracy of 0.9583 and a Kendall of 0.333. In contrast, techniques such as logistic regression and SVM were less effective in certain scenarios. This study demonstrates the potential of Machine Learning to optimize logistical processes, highlighting the importance of a detailed and careful data analysis. The findings offer significant contributions to the field of logistics, showing how the integration of advanced Machine Learning techniques can be effectively applied in an industrial context while also pointing to avenues for future improvements.
Descrição
Palavras-chave
Machine learning Logistics Osquare Imitation learning Kendall Accuracy Edit distance Route prediction Process optimization Previsão de rotas