This study investigates weather forecasting in the Mekong Delta to improve temperature prediction for agriculture and disaster management. The research addresses the limited application of advanced deep learning models, particularly iTransformer, in this context. It aims to predict hourly and daily temperatures using historical meteorological data from the Vinh Long station (2014–2024). The dataset comprises 10 years of hourly data from six western provinces in Vietnam, focusing on temperature, humidity, barometric pressure, and wind speed. The proposed methods include training various models on this data and comparing their prediction accuracy. The study evaluates traditional models (ARIMA, SARIMAX), machine learning methods (XGBoost, Prophet), deep learning models (CNNs, LSTMs, Transformers, iTransformer), and hybrid approaches (ConvLSTM, Transformer-iTransformer). The expected outcome is to identify the most effective model for enhancing weather prediction in Vietnam, with LSTM anticipated to deliver the highest accuracy, achieving the lowest RMSE. The experimental results in this work are promising, among employed models, we have RMSE of models LSTM, SARIMAX, XGBoost up to responsively 0.53, 0.649, 0.769.