Using Machine Learning (ML) Methods to Predict Temperatures in Iraq (Baghdad)

Abstract

Temperature forecasting is an important tool for understanding climate patterns and the process of adapting societies to climate change. Predictive models play a role in mediating the decision-making process in various fields. In the present study, we used three methods of machine learning - support machine learning (SVM) and artificial neural networks (ANN) and Extreme gradient boosting )XGBoost( regression - to predict the temperature in Baghdad province. The results showed the predictive ability of these methods, while the XGBoost model was the most effective, with the lowest mean square error (MSE) of 0.42. While the MSE value of the SVM model is equal to 0.85. As for the ANN model, the MSE is greater than 1.10, which indicates the possibility of overfitting. During the employment of clustering and nested trees, the results showed that XGBoost was more suitable for the complexity of the dataset, while SVM performed well but was less accurate than XGBoost. In the end, I recommend conducting research on the features of additional data and investigating better accuracy in future predictions during model refinement.