Self-evolving multi-layer perceptron (seMLP) with its applications in trend reversals & technical trading indicators

• Main Author: Seow, Wen Jun
• Other Authors: Quek Hiok Chai
• Format: Final Year Project
• Language: English
• Published: 2019
• Subjects: DRNTU::Engineering::Computer science and engineering::Computing methodologies::Artificial intelligence
• Online Access: http://hdl.handle.net/10356/76893
• Institution: Nanyang Technological University

Artificial neural networks have increased in popularity in the recent few years especially with its success in many fields of application. As evident in recent researches, there has been growing interest in the crafting of neural network architectures using an automatic process as opposed to expert tuning to find the best architecture. However, many proposed systems to automatically tune the architecture of neural networks only work for networks with a single hidden layer. Others start off with a pre-determined architecture and proceed to prune the redundant links and nodes. Hence, these approaches suffer from: (1) A pre- determined structure at the start limits the adaptability of the network to the data, (2) difficulty in extending their approach to deep neural networks with multiple hidden layers. This report proposes a novel self-evolving neural network system called self-evolving Multi-Layer Perceptron (seMLP) which can overcome the above issues in order to achieve a neural network architecture that can be determined without expert tuning and can represent abstraction of the data. seMLP applies the human cognitive ability of concept abstraction into the architecture of the neural network. The genetic algorithm in seMLP is able to determine the best architecture of a neural network that is capable of knowledge abstraction of the data. After determining the architecture of the neural network with the minimum width, seMLP is capable of pruning the network subsequently to remove the redundant links in the network thus decreasing the density of the network and achieving conciseness. The predictions from seMLP has been used in technical trading indicator such as the MACD and RSI indicator to decrease the time lag inherent in moving average windows. seMLP shows particularly strong results for the MACD indicator when modelling stock index prices that allows an investor to earn a profit above that of using a normal exponential moving average window. The results have been encouraging for the moving averages with a prediction window.