Abstract:
|
With recent advances in machine learning algorithms and the growing popularity of Big Data, the emphasis of research has shifted to become more data-centric. As a result, improving the feature selection technique is regarded as a useful method of increasing the overall accuracy of a machine learning model. Feature selection is an important part of machine learning because it involves choosing the best features to train the machine learning model. This is especially useful because it reduces noise in the input data, allowing the algorithm to train faster while also reducing the complexity of the model and making it more interpretable. Furthermore, it may result in a significant increase in model accuracy and allows for better generalization. In this talk, we propose a filter method for feature selection in binary classification problems. The method is based on a Bayesian analogue of the two-sample t-test that employs relative belief ratio. The applicability of the proposed technique will be illustrated through several real life data.
|