ACTIVATION FUNCTION IN LSTM FOR IMPROVED FORECASTING OF CLOSING NATURAL GAS STOCK PRICES

Abstract

The closing price of natural gas stocks greatly influences investment decisions and the energy industry. Predicting prices correctly can greatly help investors, market participants, and all parties involved, as it allows for making better decisions and optimizing investment portfolios. By using deep learning methods to role model various LSTM activation functions, such as Sigmoid, ReLU, and Tanh, this exploration will hopefully help understand complex patterns in time series data. By finding an appropriate forecasting method, all parties involved can reduce the environmental impact. The experimental results show that the model with ReLU activation function has the highest R2 value of 0.960 in both the training and test sets, and the model with Tanh activation function is also successful, with R2 values of 0.950 in the training set and 0.949 in the test set, and an MSE of 0.002. The model with the sigmoid activation function was slightly lower, with R2 values of 0.931 in the training set and 0.943 in the test set, and an MSE of 0.003. These findings indicate that the LSTM model with the ReLU activation function is considered better for predicting the closing price of natural gas stocks. These findings may help investors, stakeholders, and market participants choose the most accurate model to predict the closing price of natural gas stocks.