| dc.description.abstract | This thesis aims to determine whether machine learning algorithms effectively predict default loans and which machine learning algorithms are better performers than others. The research started by gathering information about which machine learning methods there are, implementing and processing their algorithms, and determining their performances.
For this thesis, I have used a dataset from a Taiwanese credit lending company consisting of 30,000 credit lenders, whereas the defaulters of this dataset are known. The choice has been made to train, test and validate six different machine learning algorithms, determine their performances, and gather helpful information on whether they are accurate in their predictions or flawed.
The main research question for this thesis is:
How effective are machine learning algorithms in predicting defaults in loans?
Some model performance measures have been used to determine the machine learning algorithms’ performances. The Area Under the Curve has been set as a primary model performance measure. In this classifier measure, a score between 0 and 1 is calculated. While a classifier with 1 AUC is the perfect model, a classifier with 0.5 AUC is as good as a random guessing one. There are also three other measurements to determine the final models: Recall, Precision, and Accuracy. The table below showcases the performance of the six algorithms after training and validation on the original dataset | en_US |
