Stock price prediction: a comparative analysis of classical and quantum neural networks

Abstract

Stock price prediction is one of the fast-growing fields, in developing countries like Tanzania it has a significant impact on investments in the public sector, private, and individual investors. The traditional prediction methods under statistical and econometric techniques cannot handle non-stationary financial time series stock data and produce accurate results. Hence, machine learning techniques are used for stock price prediction for this complex problem relying on past stock prices. This study utilizes a Long Short Term Memory neural network designed with a dropout layer, which reduces data's overfitting and diffusion gradient by Rectifier Linear Unit activation. Further, a Multiple Linear Regression network with more hidden layers for additional tuning of weights for better prediction and a Variational Quantum Neural Network utilizing PennyLane and Quantum Dense libraries to run quantum circuits for prediction are considered. The results of performance predictors like the coefficient of determination, the mean square error, the mean absolute error, and the mean absolute percentage error are compared to identify the best neural network for low stock price prediction on the National Microfinance Bank and the Cooperation Rural Development Bank data of Tanzania.