An Application Of Machine Learning Methods On Complicated Large Scale Dataset

 Machine learning research has been an upcoming trend over the last few years. With more computational power and increasing volume of data available thanks to the development of the Internet, machine learning methods could be applied to real life problems and produce fascinating outcomes. Furthermore, with the rise of deep learning methodologies, machine learning practitioners can work on unstructured datasets and achieve human level accuracy. The present thesis focuses on a structured dataset with unstructured fields and information, aiming to apply multiple machine learning methods from a supervised learning perspective. Firstly, linear regression models and extreme gradient boosting machine, as conventional machine learning methods, are applied on certain selected features of the dataset, achieving remarkable results. They serve as base models. Next, an ensembled neural network model with four modules, i.e., fully connected module, embedding module, convolutional network module, and recurrent network module, is created and implemented with the transfer learning technique, which yields better outcomes. This work leads to a typical supervised learning structure and can be considered as a stepping stone for similar practical implementations.