Hyperparameter Tuning for LSTM and ARIMA Time Series Model: A Comparative Study

Numerous forecasting issues involve time, that's why time series forecasting is an essential component of machine learning. Since computer resources have gotten better, it is now possible to record and analyse a lot of data. This opens up a lot of great chances (in many fields) to learn useful things and help the growth. In the past, there have been a lot of ways to accurately predict the next lag of a time series. When predicting future lags of time series, the ARIMA model has been shown to be more precise and accurate than other statistical models. Deep Learning algorithms are now used for time series forecasting thanks to recent improvements in the power of computers to do maths. Because it can remember past sequences, the LSTM model is one of them. How well LSTM and ARIMA models work depends on how the hyperparameters are set. In this study, Grid Search, Randomised Search, and Bayesian Search are used to tune hyperparameter to find the best set of it. These algorithms are contrasted so that the optimal model can be constructed. Consequently, Bayesian Optimisation yields the finest results.