Short - Term Temperature Forecasting Using LSTM -CatBoost Combination Method

The Internet of Things (IoT) is a fast-growing technology that has the potential to revolutionize various aspects of our daily life, such as weather forecasting. Accurate weather forecasting allows people to prepare for sudden weather changes, reducing property damage caused by weather disasters. Traditional long short-term memory (LSTM) is good at modeling long-term dependencies, but it has the problem of overfitting. In view of this, categorical boosting (CatBoost), a machine learning model having strong forecasting ability and not easy to overfit, is employed to enhance weather forecasting accuracy. By combining these two powerful and heterogeneous models, an LSTM-CatBoost averaging method, which has both low variance error and bias error in principle, is proposed to forecast weather-related variables, e.g., temperature. This study uses the data from the IoT-based automatic weather station deployed across Botswana to conduct experiments, and the results show that the proposed LSTM-CatBoost averaging method outperforms other benchmark models, including the individual LSTM and CatBoost prediction model. This indicates that the proposed method is more suitable for short-term weather forecasting in relevant regions of Botswana than some existing mainstream methods.