Time Series Prediction of Temperature Using Seasonal ARIMA and LSTM Models

Abstract

Precise quantitative understanding and monitoring of temperature is indispensable due to its tremendous impact on almost every aspect of our lives. This work investigates prediction capabilities of two machine learning techniques, namely Seasonal Autoregressive Integrated Moving Average (SARIMA) and Long Short-Term Memory (LSTM) and compares them in predicting monthly mean temperature time series data for a weather station in Ankara, Türkiye from January 2010 to March 2023. The comparison of forecasting performance was based on mean square error (MSE), root mean square error (RMSE) and mean absolute error (MAE). The results showed that both models can capture the variations of time series data. Both models exhibited reasonably good performance in predicting monthly mean temperature, but the ARIMA model gave the least forecast error compared to the LSTM model.