Michael Søegaard
Data Scientist
michael@soegaard.ai – +45 26 20 85 35
Project 8: Quant – Cointegration between two financial assets
Testing for cointegration between two financial assets, in this case, two forex pairs, and finding the optimal parameters for trading them. Cointegration is a form of pairs trading where you buy one and sell another asset to hedge your trade. By testing the pair for cointegration we know if the two assets tend to have the same price movement over time. Every time the spread between them widens we know the spread will revert to the mean. This gives us an opportunity to trade the two assets. In this project, I test multiple assets to see if they are cointegrated.
Project 7: Azure ML engineer – Capstone project
For my capstone project as an Azure ML engineer. I build a pipeline utilizing an AutoML model for baseline as well as an XGBoost model finetuned using Hyperdrive. Custom docker environments were created and the resulting model was deployed and consumed.
Dataset for the project is from the Kaggle dataset, “Health Insurance cross sell”.
To read more and get the code, click the headline or github logo.
Project 6: AzureML – Comparing Hyperdrive vs AutoML
Project 5: Flight delay project – Kaggle dataset
Data based on this Kaggle dataset. We get data from January 2019 and 2020 and need to prdict flight delays. In this project there is a lot of work in preprocessing. A lot of missing values and different categorical features. A thorough analysis and comprison of various algoriths such as Logistic regression, XGBoost, Random Forest etc. To read more and get the code, click the headline or github logo.
Project 4: Forex robot using reinforcement learning
A pretty complex, but very exciting project. In this project I wrote a Double Dueling DQN program to trade the pound/Yen forex pair. It involves a lot of hyperparameters and many choices to be made. A lot of challenges was met and conquered, and a lot of coffee got consumed. 😀 This model was tested in live trading. To read more and get to the code, click the headline or Github logo.
Project 3: XGBoost model conquers the forex challenge
A spin-off from the above project. When learning about decision trees, I decided to try and use XGBoost on forex trading. A simpler approach than the one above. I used the same dataset and features as in the reinforcement version, but used a regression value as target. Meaning, I wanted it to forecast at what price a program should take profit given the current sample. This model was also tested in live trading. To read more and get to the code, click the headline or Github logo.
Project 2: Audiobooks – predict customer churn
Project 1: Keras LSTM model – tuning using Keras tuner
