Deep neural network with fuzzy inputs for portfolio management

Abstract

Deep learning is a type of machine learning that attempts to simulate the learning behavior of our human brain. Such model with neural networks that consist of multiple layers can learn from large amount of data. Despite the capabilities of deep learning, the learning process itself is still a black box. As researchers and even just as end users of deep learning, we are curious to find out the learning process that takes place in the black box instead of blindly trusting the outcome it produces. This dissertation explores an Interpretable Fuzzy Neural Network that allows us to better understand the learning process that takes place within the black box through fuzzy logic in the intelligent system. The model will be applied algorithmic finance to firstly predict future stock prices, and by extension predict trend reversals using technical indicators. The performance of the model will be evaluated through back testing and comparing against the performance of a benchmark.