Machine Learning Final Project

The goal of this project was two fold. The first goal was to investigate how hyperparameter manipulation impacted the functionality of different classic machine learning algorithms in a binary and ternary classification stock trading strategy. The results of this project demonstrate not only the necessity for hyperparameter optimization in financial data sets, but also the sensitivity that different algorithms have to classification strategies. We used features generated by rolling averages of closing price for a set of highly liquid stocks (very common in momentum based strategies) and the targets were next day returns (after accounting for trade friction) classified as either a 1, 0, or -1 if the next day returns were positive, neutral, or negative, respectively.

If you are interested in the second part of our project, feel free to email me!

What is the Significance? This project demonstrates the necessity for optimizing hyperparameters and choosing a classification strategy that pairs well with a particular machine learning strategy. Our work shows that we were able to outperform a simple holding strategy for certain stocks over a particular timeline by optimizing hyperparameters and choosing the optimal pair of machine learning algorithms with classification strategies. This work could be demonstrated more concretely by expanding the data set and considering more classification strategies.

• Verizon Data

• Advanced Micro Devices Data

• Nasdaq Data

• Apple Data

**This was a group project**

• Machine Learning Project Code (Part1)

• Machine Learning Project Code (Part2)

• Machine Learning Project Code (Part2)

• Machine Learning Project Code (Part2)

• Machine Learning Project Code (Part2)