Machine Learning Project

Machine Learning course 2021-2022.
Masters' Degree in Applied Mathematics, Sapienza University of Rome.
Exam date: June 13th 2022

Dataset and task description:

The provided dataset consists of daily MSFT financial data from October 2013 to February 2022 ¹. These data are initially pre-processed, then used to define some prediction models. Given day $i$, consider the price of the stock in the following 10 days $i+1,...i+10$.
Our target is to predict the highest price reached by MSFT stock in each of these 10 days, given information on day $i$.
See this notebook for detailed analysis and results.

Feature engineering

Some fetures of the dataset are highly correlated to each other, and so redundant, since they don't provide the model with any additional useful information.
However, new input features (technical indicators) can be created from these correlated features.
Here the following indicators have been implemented:

1. Simple moving average.

2. MACD (Moving Average Convergence/Divergence).

3. Average true range

4. Average typical price.

5. CCI (Commodity Channel Index)

Model selection and evaluation

After feature selection and data transformation are performed, the following models are trained and evaluated by spitting the dataset for crossvalidation:

• Linear regressor
• Multilayer perceptron
• Regression trees
• Random forest

Footnotes

1. Data have been downloaded from https://www.marketwatch.com/investing/stock/msft/download-data. ↩