Comparative Analysis of ARIMA and LSTM Models for Weather Forecasting Using Time Series Data

Time series forecasting plays a crucial role across various domains, including transportation planning, supply chain management, and environmental sciences. Traditional statistical approaches such as Autoregressive Integrated Moving Average (ARIMA) have long been favored for their robustness and interpretability in modeling univariate time-dependent data. However, with the advent of increased computational power and advancements in machine learning, particularly deep learning, novel algorithms like Long Short-Term Memory (LSTM) networks have emerged, offering the ability to capture complex temporal dependencies through their specialized memory gates. This paper conducts a comparative analysis of ARIMA and LSTM models applied to real-world temperature datasets from two distinct climatic regions (Budapest and Los Angeles). Both models were evaluated using standard performance metrics to assess their efficacy in short-term temperature prediction. The findings highlight the relative strengths and weaknesses of each approach, with LSTM demonstrating notable improvements in forecasting accuracy over ARIMA, albeit with considerations regarding training complexity and resource requirements. These results contribute to a better understanding of when and how deep learning models can be effectively employed in practical forecasting tasks alongside traditional methods.