Predicting stock trend using an integrated term frequency–inverse document frequency-based feature weight matrix with neural networks

Abstract

The financial market consists of various money-making strategies wherein trading through a stock market is an important example. The complex non-linear behaviors of volatile stock markets attract researchers to study inherent patterns. As the primary motivation for investment in such markets is to gain higher profits, potential stocks are given considerable attention using various weighting strategies that can enhance future returns. Term frequency–inverse document frequency (TF–IDF) is a statistical approach with remarkable applications in the financial domain for information retrieval from textual data; it identifies the importance of a term in the given document of a corpus. However, the application of TF–IDF for the numerical data representation is explored to a limited extent. In this article, we propose to extend the applicability of TF–IDF for the numerical time-series stock market data; we process the data and prepare them to be suitable for TF–IDF. We utilize this statistical approach to derive feature weight matrix from the historical stock market data and further, integrate it with the widely explored neural network architectures namely, backpropagation neural network (BPNN), long short-term memory (LSTM), and gated recurrent unit (GRU) for predicting stock market trend. Simulation results show that the proposed integrated approach using TF–IDF-based feature weight matrix and neural networks outperforms the considered recent approaches. The results are statistically supported with p-value less than .01 using a Wilcoxon signed-rank test; our proposed approach is supported with illustrative examples to develop better understanding of the work. Also, remarks on the conclusions and potential future scope are discussed.