Generic Architecture for Predictive Computational Modelling with Application to Financial Data Analysis: Integration of Semantic Approach and Machine Learning

The PhD thesis introduces a Generic Architecture for Predictive Computational Modelling capable of automating analytical conclusions regarding quantitative data structured as a data frame. The model involves heterogeneous data mining based on a semantic approach, graph-based methods (ontology, knowledge graphs, graph databases) and advanced machine learning methods. The main focus of my research is data pre-processing aimed at a more efficient selection of input features to the computational model.

Since the model I propose is generic, it can be applied for data mining of all quantitative datasets (containing two-dimensional, size-mutable, heterogeneous tabular data); however, it is best suitable for highly interconnected data.

To adapt this generic model to a specific use case, an Ontology as the formal conceptual representation for the relevant domain knowledge is needed.

I have determined to use financial/market data for my use cases. In the course of practical experiments, the effectiveness of the PCM model application for the UK companies’ financial risk analysis and the FTSE100 market index forecasting was evaluated. The tests confirmed that the PCM model has more accurate outcomes than stand-alone traditional machine learning methods.

By critically evaluating this architecture, I proved its validity and suggested directions for future research.