Global Temperature Predictions Using Auto-ARIMA Model

This research focuses on global temperature forecasting using the Auto-ARIMA model. The study utilizes month-by-month historical temperature data from 1961 to 2019, sourced from Kaggle, to create accurate predictions for selected continents, sub-regions, and countries, along with a global prediction for all twelve months. The data was trained and tested on the same dataset using an 80-20 split to ensure enough reserve data for training. The research focuses on the change in temperature by month. The Auto-ARIMA model was chosen due to its high efficiency in time-series forecasting, as it dynamically adjusts parameters to generate optimal predictions. By integrating the ARIMA model within Google Colab, this research ensures accessibility and computational efficiency due to the software’s cloud capabilities, providing access to large computational power. The model's performance for the summer months was evaluated using Root Mean Squared Error (RMSE), Bayesian Information Criterion (BIC), and Akaike Information Criterion (AIC) highlighting its capability to predict temperature trends accurately. The model made predictions with a Log Likelihood of 10.319, an RMSE of 0.288, an Akaike Information Criterion (AIC) of 1.363, and a Bayesian Information Criterion (BIC) of 5.739. The objective of the project was to minimize the RMSE value to maximize prediction accuracy. Challenges such as unsupported index warnings and non-stationary parameters were encountered and addressed, emphasizing the need for further optimization in climate modeling. The analysis provides actionable insights into global and regional temperature variations, emphasizing the importance of accurate forecasting in addressing climate change. The simplicity and transparency of the information make it easier for policymakers to make informed decisions based on predictions resulting in relevant and effective environmental policy. Advanced visualization techniques combined with statistical models improve predictive modeling comprehension while demonstrating its practical application to environmental policy development.