| Name: | Description: | Size: | Format: | |
|---|---|---|---|---|
| DM_BernardoAlmeida_MEI_2022 | 10.57 MB | Adobe PDF |
Advisor(s)
Abstract(s)
Nos últimos anos, tem havido um crescimento na utilização de Machine Learning e uma necessidade
crescente de aplicar modelos de Machine Learning a várias necessidades empresariais, desde a análise
dos padrões de compra dos clientes até à tomada de uma decisão empresarial para fazer crescer esse
mesmo negócio.
Num ambiente empresarial acelarado que nos encontramos atualmente, desenvolver e disponibilizar
um bom modelo pode não ser um processo muito célere. O principal motivo são os dados necessários
para obter o bom modelo, visto que para obtê-lo pode ser necessário uma grande quantidade de dados
e isto pode afetar o tempo de treino do modelo, ou pode ser necessário um pré-processamento dos
dados, levando ao aumento do tempo para obter o bom modelo. Com isto, este trabalho apresenta
uma possível solução para este problema, onde, através do Active Learning, o humano aplica etiquetas
a uma pequena quantidade dados, de seguida são criados vários modelos com parâmetros diferentes
para serem treinados até que um intervalo de valores seja atingido. Por fim, algumas métricas serão
extraídas e analisadas para concluir qual o melhor modelo. Por fim é apresentada a previsão do
modelo em conjunto com uma explicação com o que o modelo considerou importante.
In recent years, there has been a growth in the use of Machine Learning and an increasing need to apply Machine Learning models to various business needs, from analysing customer buying patterns to making a business decision to grow that same business. In the fast-paced business environment we currently find ourselves in, developing and delivering a good model may not be a very fast process. The main reason is the data required to obtain the good model, since to obtain it may require a large amount of data and this may affect the training time of the model, or a pre-processing of the data may be required, leading to increased time to obtain the good model. With this, this work presents a possible solution to this problem, where, through Active Learning, the human applies labels to a small amount of data, then several models are created with different parameters to be trained until a range of values is reached. Finally, some metrics will be extracted and analysed to conclude which model is the best. Finally the prediction of the model is presented together with an explanation of what the model considered important.
In recent years, there has been a growth in the use of Machine Learning and an increasing need to apply Machine Learning models to various business needs, from analysing customer buying patterns to making a business decision to grow that same business. In the fast-paced business environment we currently find ourselves in, developing and delivering a good model may not be a very fast process. The main reason is the data required to obtain the good model, since to obtain it may require a large amount of data and this may affect the training time of the model, or a pre-processing of the data may be required, leading to increased time to obtain the good model. With this, this work presents a possible solution to this problem, where, through Active Learning, the human applies labels to a small amount of data, then several models are created with different parameters to be trained until a range of values is reached. Finally, some metrics will be extracted and analysed to conclude which model is the best. Finally the prediction of the model is presented together with an explanation of what the model considered important.
Description
Keywords
Active Learning Explainable AI Labelling Early Stopping