Stock Price Prediction using ResNLS Technique

Forecasting stock prices have never been an easy task. While many attempts employ deep learning and machine learning techniques to deal with the problem, few take into account the various degrees of associations among stock prices. This paper provides a hybrid approach that highlights the relationships between related stock prices to enhance stock price prediction. The two neural networks that make up the proposed approach, ResNLS, are ResNet and LSTM. Although LSTM analyzes the original time-series data using a collection of interdependence that are treated as residuals. ResNet acts as a feature extractor to find relationships between stock prices over time frames. Our research shows that the model (ResNLS-5) performs best while the SSE Composite Indexing will be calculated using the close price data from previous five days of trading. This is in difference from models that utilize different inputs. In the context of accuracy for forecasting, ResNLS-5 performs better than the standard BiLSTM, LSTM, RNN, and CNN models. Furthermore, it shows at least an increase of 20% over the existing standards as of right now. We build a quantitative trading system for back testing in order to confirm if ResNLS-5 can assist customers in profitably avoiding risks and making gains in the stock market. The outcomes of the experiment demonstrate that a trading approach derived from ResNLS-5 forecasts may effectively reduce losses in periods when stock values are declining and increase profits in periods when stock values are inclining.