Information fusion-based genetic algorithm with long short-term memory for stock price and trend prediction

Abstract

Information fusion is one of the critical aspects in diverse fields of applications; while the collected data may provide certain perspectives, a fusion of such data can be a useful way of exploring, expanding, enhancing, and extracting meaningful information for a better organization of the targeted domain. A nature-inspired evolutionary approach, namely, genetic algorithm (GA) is adopted for a variety of applications including stock market prediction. The complex, highly fluctuating financial market-related problems require optimized models for reliable forecasting. Also, it can be observed that stock market etiquettes are generally non-linear in nature and therefore, a broader understanding and analysis of such market behaviors necessitate the collection and fusion of relevant information based on different associated factors. In this article, we propose an information fusion-based GA approach with inter-intra crossover and adaptive mutation (ICAN) for stock price and trend prediction. Inspired by the genetic diversity and survival capability of various organisms, our proposed approach aims to optimize parameters of a long short-term memory prediction model, and selects a set of features; to address these problems of interest, we integrate inter-chromosome as well as conditional intra-chromosome crossover operations along with adaptive mutation to diversify the potential chromosome solutions. We illustrate the step-by-step procedure followed by GA with ICAN and evaluate its performance for one-day-ahead stock price and trend prediction. GA with ICAN-based optimization results in an average reduction of 43%, 27%, and 26% using mean squared error, mean absolute error, and mean absolute percentage error, respectively, as compared to the existing GA-based optimization approaches; further, an average improvement of 61% is encountered using R2 score. We also compare our work with Ant Lion Optimization approach and demonstrate the significance of GA with ICAN-based optimization. We analyze statistical significance, as well as convergence functions, for GA with ICAN and discuss remarkable performance enhancement; we provide necessary concluding remarks with potential future research directions.