Please use this identifier to cite or link to this item:
|Title:||Fuzzy-Embedded Gated Recurrent Unit (FE-GRU) with application in stock trading||Authors:||Lim, Chee Yuan||Keywords:||Engineering::Computer science and engineering||Issue Date:||2022||Publisher:||Nanyang Technological University||Source:||Lim, C. Y. (2022). Fuzzy-Embedded Gated Recurrent Unit (FE-GRU) with application in stock trading. Final Year Project (FYP), Nanyang Technological University, Singapore. https://hdl.handle.net/10356/157227||Project:||SCSE21-0440||Abstract:||Artificial intelligence is a field in computing that aims to embed intelligence and thinking into machines to help humans accomplish tasks more effectively and efficiently. Neural network was later made popular when computer scientists started studying the process of human thinking, as well as the physical operations of human brain. Since then, the use of neural network and artificial intelligence have been adopted in many fields including Finance to detect fraud or trend analysis, Engineering to predict machine failure or replacement rates, Healthcare to propose optimal treatment solutions, Architecture to analyse safety of building structure and many more. This dissertation proposes an architecture known as Fuzzy Embedded Gated Recurrent Unit (FE-GRU) which embeds both fuzzy and recurrent neural network together to explain the black box behaviour of a neural network. This network will be tested using time sensitive (sequential) data and more explanations will be provided to allow better understanding of the FE-GRU. Although this paper explores the use of a combination of fuzzy neural network and recurrent neural network, the evaluation will be done based on trend indicators, and whether there is possibly a better combination of trend indicators. From the raw data obtained from Yahoo Finance, scaling and fuzzification will be performed before it is fed into the GRU as mentioned before. The GRU neural network seeks to determine sequential trends to predict respective outputs and feeds information by tagging respective rule nodes required in the fuzzy neural network. Upon defuzzifying outputs, they are then compared against the actual close price. The performance of the FE-GRU is evaluated and compared against the ARIMA model, vanilla GRU model using Apple’s stock data. Both the ARIMA and vanilla GRU model returned R2 value of 0.98 and 0.89 respectively. The FE-GRU on the other hand have returned a R2 of 0.94. Although the RE-GRU failed to perform better in this case, it still managed to provide profitable returns when used with trend indicators. The conclusion of the trend indicators was that there is no single best indicator and no one best combination of indicator and this will be further explained in subsequent Chapters.||URI:||https://hdl.handle.net/10356/157227||Fulltext Permission:||restricted||Fulltext Availability:||With Fulltext|
|Appears in Collections:||SCSE Student Reports (FYP/IA/PA/PI)|
Updated on May 20, 2022
Updated on May 20, 2022
Items in DR-NTU are protected by copyright, with all rights reserved, unless otherwise indicated.