A framework for the analysis of discrete irregular patterned sequential datasets

Abstract

Trial-and-error is often used for developing new deep learning models. There is no systematic procedure for improving existing models for predicting irregular sequential time-series. This research developed a systematic framework to improve state-of-the-art deep learning models for financial time-series prediction. The framework was used to create an enhanced model. A design science research methodology was used to design this framework with customised multidimensional evaluation criteria and metrics. The model was applied to predict currency exchange rates more accurately than existing state-of-the-art models found in the literature. The main contribution of this work is a framework that provides a procedure for improving deep learning models. As a proof of its usefulness, the framework was applied to develop an improved model.