Adaptive trio-ensemble deep neural network for high-frequency stock price prediction

Abstract The analysis and forecasting of stock price is a highly complex task since its inception. Researchers have proposed a hundreds of mathematical and machine learning based models to solve this high frequency prediction problem. The constraints that restricts the effective stock market forecasting method is its dependency on variety of factors like news, announcement of dividends, company policy, drastic changes at management level, launch of new products etc. The characteristics of Deep learning algorithms like choice of network structure, activation function, and other model parameters etc voted it as a best choice for prediction. This paper proposed an ensemble prediction model by exploiting three most promising variant of Deep Neural Network (DNN) namely Gaussian, Poisson, and Gamma out of six available probability distributions (Quantile, Gaussian, Poisson, Laplace, Huber, and Gamma). The experimental results show that the proposed ensemble deep learning model claimed the best accuracy of R2: 0.92 and Root Mean Square Error (RMSE): 0.17 as per the literature reviewed in this category.