An evolving type-2 neural fuzzy inference system with fuzzy rule interpolation (eT2FIS++) with its application in straddle option trading

Fuzzy neural networks are often used for modelling dynamic data streams and the systems keep evolving from offline to online, innovating and adding new schemes to address each individual issue of sparsity, non-linearity and time-variants in the datasets. The research has been widely applied to different areas such as traffic control, flood or rain prediction and financial worlds. In particular, it is topical to model the data in financial markets. However, many existing systems are incapable of handling sparse and dynamic time series data streams such as option trading data in the financial markets. Interpolation and extrapolation are one of the most popular techniques in handling the sparsity in the datasets. Inspired by the research by Huang [76] and Chen [91], this paper extends the established work of Tung [90] with interpolation/extrapolation technique. This equips the existing system from Tung [90] with the ability of handling sparse data and invoking interpolation when concept drift or shift is detected. The proposed model is named as Evolving Type-2 Neural Fuzzy Inference System with Fuzzy Rule Interpolation (eT2FIS++). Inherited the properties of eT2FIS, eT2FIS++ has the following advantages: 1) it is an incremental learning system; 2) it has the known noise resistance capability; 3) it ensures a compact and up-to-date rule base; 4) it is able to handle concept drift via interpolation even in sparse environment. The proposed eT2FIS++ model is benchmarked against several models by using datasets with different properties. It is then deployed in an intelligent trading system that is used for option straddle trading. The results are very encouraging.Bachelor of Engineering (Computer Science