Stock trading & prediction using deep learning neural networks

• Main Author: Gowri, Kannan
• Other Authors: Wang Lipo
• Format: Final Year Project
• Language: English
• Published: 2017
• Subjects: DRNTU::Engineering::Electrical and electronic engineering
• Online Access: http://hdl.handle.net/10356/71631
• Institution: Nanyang Technological University

This report describes various deep neural network models based on technical analysis to predict stock prices. Three stocks – Walmart, HSBC and Petrobras are chosen to test Multilayer Perceptron, Deep Belief Network (DBN), Convolutional Neural Network (CNN) and Random Forests. A hybrid model comprising of a DBN with a NN stacked on top is proposed in this report. 25% of the data is used for testing while the rest is used for training and validation of models. The models are tested for a wide range of parameters including activation functions, number of hidden layers, 5 day vs 1 day forecast, 20, 10 and 5 day prior input (batch size), epochs, learning rate and momentum. The prediction results for the model with the optimal set of parameters are compared against a published paper’s work on Multilayer Perceptron model that used the same input data sets.