Interpretable fuzzy deep neural system for stock price modelling with applications in algorithmic finance

• Main Author: Chai, Alwin Wei Heng
• Other Authors: Quek Hiok Chai
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2024
• Subjects: Computer and Information Science
• Online Access: https://hdl.handle.net/10356/176598
• Institution: Nanyang Technological University

With the increasing popularity of artificial intelligence, deep learning via deep neural network architectures are increasingly utilized for regression and classification tasks and in recent history for the generation of content. Deep learning offers immense opportunity for organizations and individuals to leverage off its ability to learn complex patterns otherwise less possible using conventional models. In the financial domain, there is much profit to tap on the intelligence of deep networks, to segment sectors of growth, predict future values and to manage financial resources in a data-driven manner. However, the usage of deep networks typically confers less explanatory power, which can be unsettling when the stakes are high. In this project, a novel neuro-fuzzy system, the Interpretable Fuzzy Deep Transformer Neural Network (IFDTNN) proposed attempts to improve interpretability to deep neural network systems where fuzzy rule mappings implemented through the Mamdani fuzzy inference system can be reconciled with human knowledge. The IFDTNN engineered for financial modelling, serves as the apt architecture for financial use featuring both predictive abilities and interpretability to convince stakeholders with interpretations, potentially providing decision rationale. Leveraging on the predictive abilities of the IFDTNN, lag caused by technical indicators can be better managed for improved trading performance, maximizing returns.