An optimal limit order book prediction analysis based on deep learning and pigeon-inspired optimizer

Abstract

Currently, In the system-driven electronic markets, stocks and futures contracts are traded based on the centralization of buy and sell orders in the boundary order book (LOB), as provides much more information about stocks than their prices, such as dynamics of the price and predictability of the next trading move. This paper proposes an optimal limit order book analysis for five stocks, including Amazon (AMZN), Apple (AAPL), Google (GOOG), Intel (INTC), and Microsoft (MSFT). It is based on deep learning and an enhanced pigeon-inspired (PIO) algorithm. The system reduces the dimentaionlty of LOB data sets by using a pigeon-inspired optimizer to determine the most significant features. The fitness function is used to evaluate the fitness value of each solution based on TPR and FPR and the feature count. The optimized LOB feature selection is evaluated using the Decision Tree (DT) classifier. A new deep neural network with high-frequency order series includes convolutional, dense layers and Inception units to predict future stock price movements (Submission, Cancellation, Deletion, Execution visible and hidden orders) in an extensive high-frequency LOB database that supports improving the operational performance of the trading process. The proposed model is evaluated using the LOB dataset, and the results show that the model performs better in predicting the different classes. The analysis of variance ANOVA supports the obtained results that test for the significant difference among the means of all items according to their event types.