Long Short-Term Memory Network Hyperparameter Optimization using Hybrid Algorithm GA-PSO on LQ45 Stock Prediction

Abstract

Stock is a good investment tool, keeping money from inflation, and very trendy to earn a living nowadays by becoming a trader. There is always a risk, especially when trading, because stocks can fluctuate easily depending on the company. One of the data science capabilities, prediction modeling, can help lower the risk by predicting the stock price movement. This research proposed a prediction sequential data model, an optimized hyperparameter LSTM Network using hybrid GA-PSO (LSTM-GA-PSO). Hybrid GA-PSO aims to overcome the GA problem in terms of slow execution time and PSO that tend to be trapped in the local optimum. With the characteristics of both algorithms, the hybrid algorithm can solve each other algorithms downside. The low fluctuation stock of the Indonesian Index LQ45 dataset will be used to train and test the model and compare the proposed model with LSTM-GA and LSTM-PSO. Experiment results show that the hybrid LSTM-GA-PSO has a promising performance. Hybrid GA-PSO improved 18.18% of its time execution to GA and 29.07% accuracy to PSO.