_

Abstract

The first problem of data scientists is to get enough features. The second, probably more important but also more problematic, is to use the right features or to use features right.

Using an appropriate subset of features may improve the performance of the model (e.g., accuracy during classification). It can at least speed up the run of the model what can be beneficial in finance predictions, for example.

Another approach of feature reduction is extraction. There are linear and non-linear functions that lead to smaller feature space as well as improved performance. Some of them are quite interesting because they are based on deep learning [1].

However, is it useful for price prediction? Is any of these approaches preferable? Is any more appropriate in general? We will try to find the answers during the lecture (or at least we will discuss them).

Readings

1. On Feature Reduction using Deep Learning for Trend Prediction in Finance: https://arxiv.org/pdf/1704.03205.pdf