The stock market is one of the main fields that investors are engaged in, so stock market price trend prediction is always a hot topic for researchers in financial and technical domains. In this research, we aim to build a state-of-the-art forecasting model for price trend forecasting, focusing on short-term price trend forecasting.
as fame concludes in , forecasting financial time series is known to be a notoriously difficult task due to the generally accepted market efficiency semi-strong form and high noise level. in 2003, wang et al. in  already applied artificial neural networks in stock market price prediction and focused on volume, as a specific characteristic of the stock market. one of their key findings was that volume was not effective in improving forecasting performance on the datasets they used, which was s&p 500 and dji. ince and trafalis in  aimed at short-term forecasting and applied the support vector machine (svm) model in stock price prediction. his main contribution is to make a comparison between multilayer perceptron (mlp) and svm and then found that most of the scenarios svm outperformed mlp, while the result was also affected by different business strategies. meanwhile, researchers in the financial domains applied conventional statistical methods and signal processing techniques to analyze stock market data.
Reading: How to predict the stock market
See also : How to Buy Blockbuster Stock – Benzinga
Optimization techniques, such as Principal Component Analysis (PCA), have also been applied in short-term stock price forecasting . Over the years, researchers have not only focused on stock price-related analysis, but also attempted to analyze stock market transactions such as volume explosion risks, broadening the research domain of analysis. of the stock market and indicates that this research domain still has great potential  . As AI techniques evolved in recent years, many proposed solutions attempted to combine machine learning and deep learning techniques based on previous approaches, and then proposed new metrics that serve as training features, such as liu and wang  . This kind of earlier work belongs to the domain of feature engineering and can be considered as the inspiration for the feature extension ideas in our research. liu et al. in  proposed a convolutional neural network (cnn), as well as a model based on a short-term memory (lstm) neural network to analyze different quantitative strategies in the stock markets. cnn serves for stock picking strategy, it automatically extracts features based on quantitative data and then follows an lstm to retain time series features to enhance earnings.
The latest work also proposes a similar hybrid neural network architecture, which integrates a convolutional neural network with a bidirectional long-term memory to predict the stock index . Although researchers frequently proposed different architectures of neural network solutions, it generated further discussion on whether or not the high cost of training such models is worth it.
There are three key contributions of our work (1) a new mined and cleaned dataset (2) comprehensive feature engineering, and (3) a custom long-term memory (lstm)-based deep learning model.
We have created the dataset ourselves from the data source as an open source data api called tushare . The novelty of our proposed solution is that we proposed feature engineering along with a lean system instead of just an lstm model. we looked at the previous works and found the gaps and proposed a solution architecture with a comprehensive feature engineering procedure before training the prediction model. With the success of the feature extension method collaborating with recursive feature removal algorithms, it opens the door for many other machine learning algorithms to achieve high-precision scores for short-term price trend prediction. demonstrated the effectiveness of our proposed feature extension as feature engineering. In addition, we introduced our custom lstm model and further improved prediction scores on all evaluation metrics. the proposed solution outperformed machine learning and deep learning based models in previous similar work.
The rest of this document is organized as follows. the “related jobs survey” section describes the related jobs survey. the “the dataset” section provides details about the data we extracted from public data sources and the prepared dataset. the “methods” section presents the research problems, the methods and the design of the proposed solution. This section also includes the detailed technical design with algorithms and how the model was implemented. the “results” section presents comprehensive results and an evaluation of our proposed model, and compares it to models used in most related work. the “discussion” section provides a discussion and comparison of the results. the “conclusion” section presents the conclusion. this research work has been built on the basis of shen .