Global Temperature Prediction by Convolutional Neural Network (CNN)

This paper explores the use of convolutional neural networks or CNNs for temperature prediction, based on annual climate data from 1880 to 2023, collected by NASA to track global temperature trends. The CNN model is based on convolutional, pooling and fully connected layers. The performance of the CNN model is evaluated using the Mean Absolute Error or MAE, the Root Mean Square Error or RMSE and the Coefficient of determination or R² on the training and validation sets. In the training set, MAE indicates that the model predicts on average with a deviation of 0.0836 from the true values. The RMSE result is slightly higher than MAE. R² shows that 92.92% of the variance in the training data is explained by the model. In validation, MAE is slightly higher than in training. RMSE is also slightly higher than in training, but remains relatively low. R² on the validation set is slightly lower than that of training. In this study, the model used provides reliable predictions with small deviations between observed and predicted data. Despite the good accuracy, there is still room for improvement, including the use of additional data and the integration of more advanced learning models.