Hyperparameter Optimization of Long-Short Term Memory using Symbiotic Organism Search for Stock Prediction

Producing the best possible predictive result from long-short term memory (LSTM) requires hyperparameters to be tuned by a data scientist or researcher. A metaheuristic algorithm was used to optimize hyperparameter tuning and reduce the computational complexity to improve the manual process. Symbiotic organism search (SOS), which was introduced in 2014, is an algorithm that simulates the symbiotic interactions that organisms use to survive in an ecosystem. SOS offers an advantage over other metaheuristic algorithms in that it has fewer parameters, allowing it to avoid parameter determination errors and produce suboptimal solutions. SOS was used to optimize hyperparameter tuning in LSTM for stock prediction. The stock prices were time-series data, and LSTM has proven to be a popular method for time-series forecasting. This research employed the Indonesia composite index dataset and assessed it using root mean square error (RMSE) as a key indicator and the fitness function for the metaheuristic approach. Genetic algorithm (GA) and particle swarm optimization (PSO) were used as benchmarking algorithms in this research. The hybrid SOS-LSTM model outperformed GA-LSTM and PSO-LSTM with an RMSE of 78.799, compared to the GA-LSTM model with an RMSE of 142.663 and the PSO-LSTM model with an RMSE of 529.170.