Please contact me initially by email (D.Gorse (at) cs.ucl.ac.uk) if you are interested in these projects. You can also contact me if you have an idea of your own that applies neural computing, particle swarm optimisation, or similar metaheuristic to an area of interest to you. Past applications proposed by students have been to subjects as diverse as protein structure prediction and the classification of musical timbre.

Apologies but both the projects below have now been taken.

Neural network detection of manipulative practices in tick data - now taken (confirmed)

This project would aim to use a published methodology that has been applied successfully to stock price data to detect equivalent price manipulation practices in cryptocurrency data. As the data set is imbalanced (not many instances of manipulation compared to normal trading) the work could be enhanced by the use of a technique for the better handling of imbalanced training data, such as SMOTE.

ICOs (initial coin offerings) are the equivalent for cryptocurrencies to IPOs for companies. Investors can contribute to projects when they are at an early stage, before they are listed for trading. The trading during the first 24 hours after listing is usually very volatile. This project would use a neural net to either 1) predict the price 24 hours after listing or 2) classify whether the price would be above or below the ICO price, in either case based on a number of factors such as ICO valuation price, number of ICO contributors, ICO length, and general market conditions.