Deep learning in stock market forecasting

• Main Author: Wijaya, Michael
• Other Authors: Ponnuthurai Nagaratnam Suganthan
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2022
• Subjects: Engineering::Computer science and engineering
• Online Access: https://hdl.handle.net/10356/157860
• Institution: Nanyang Technological University

Regression in machine learning is a task of predicting continuous dependent output based on multiple independent inputs. One of the best machine learning metohd is called deep learning. Deep learning is able to achieve high performance derived from high complexity of machine learning model which is essential to accurately forecast stock market price. However, this comes with cost of high computational power and high tendency of overfitting. In other words, having more parameters in the model can easily improve the performance by solving the underfitting problem, but the model is more likely exposed to overfitting which leads the model unably to reach the best expected result. Hence, one of possible solution is to randomize and freeze some of the parameters, reducing model complexity which can possibly enhance the performance. Therefore, this project experiments on multiple deep learning models as well as randomized deep learning models: Simple Recurrent Neural Networks (RNN), Gated Recurrent Unit (GRU), Long Short Term Memory (LSTM), Convolutional Neural Networks (CNN), Random Vector Functional Link (RVFL), Ensemble Deep Random Vector Functional Link (edRVFL), Echo State Networks (ESN), Temporal Convolutional Networks (TCN). We test it on historical datasets of stock market value from 5 different companies. The result shows randomized model generally works better than non-randomized model.