MLP, CNN, LSTM and Hybrid SVM for Stock Index Forecasting Task to INDU and FTSE100

Abstract

The aim of this paper is to investigate the Support Vector Machine (SVM), the Binary Gravity Search Algorithm combined Support Vector Machine (BGSA-SVM), the Multi-layer Perceptron (MLP), the Convolution Neural Network (CNN) and the Long Short-Term Memory (LSTM) neural network, when applied to the task of forecasting and trading FTSE100 and INDU indexes from 2000 to 2018. Results show that the performance of the SVM is unstable as it displays sensitivity to inputs and parameters. The other Machine Learning models (BGSA-SVM, MLP, CNN and LSTM) outperform Buy-and-Hold and Random Walk (RW). Additionally, this paper finds no significant difference between the performance of BGSA-SVM, MLP, CNN and LSTM.