COA-HONN: Cooperative Optimization Algorithm Based Higher Order Neural Networks for Stock Forecasting

Background: A broad range of nature inspired optimization techniques are proposed in the literature and applied for stock market prediction. They performed notably differently across the stock market datasets. Higher Order Neural Network (HONN) have capacity to expand the input demonstration space with fewer trainable weights and perform high learning capabilities. Methods: This article attempts to construct a Cooperative Optimization Algorithm (COA) framework as an alternative of employing solitary algorithm. The COA considers two optimizations, i.e. a genetic algorithm and an artificial chemical reaction optimization as constituent techniques. The COA framework executes each constituent algorithm with a fraction of the whole computation time budget. It encourages interaction between them, so that they can be benefited from each other. Here, optimal model parameters of two HONNs, i.e. Pi-Sigma Neural Network (PSNN) and Functional Link Artificial Neural Network (FLANN) are searched by COA, hence forming two COA-HONN hybrid models. Results and Conclusions: The models are evaluated on forecasting daily closing prices of five real stock datasets. The experimental results confirm that the COA approach enhances the prediction accuracy over individual algorithm. We conducted the Deibold-Mariano test to check the statistical significance of the proposed models, and it was found to be significant. Hence, the proposed approach can be used as a promising tool for stock market forecasting.