LEARNING-BASED TIME SERIES MODELING FOR CLIMATE CHANGE VARIABILITY ANALYSIS

Abstract

Climate change has intensified the need for accurate analytical approaches capable of identifying long-term temperature trends and improving climate prediction. This study proposes a machine learning–based time-series modeling framework for analyzing global temperature anomaly data and forecasting climate variability. The dataset consists of monthly temperature anomalies spanning more than a century, providing a comprehensive record for examining historical climate patterns and long-term warming dynamics. Several feature engineering techniques were applied to enhance the predictive capability of the dataset, including lag variables, rolling statistical indicators, exponential moving averages, and difference transformations. These engineered features capture temporal dependencies, seasonal variations, and long-term climate trends within the time series. Multiple predictive models were implemented to evaluate forecasting performance, including ARIMA, Random Forest, Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), and Transformer architectures. Model performance was assessed using standard evaluation metrics such as Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and the coefficient of determination (R²). The results demonstrate that deep learning models, particularly Transformer and GRU architectures, outperform traditional statistical methods in capturing complex temporal relationships in climate data. The findings confirm the effectiveness of advanced machine learning techniques for modeling climate variability and improving predictive accuracy. This research contributes to climate data analysis by providing a comprehensive framework that integrates feature engineering and comparative machine learning modeling for long-term temperature forecasting.