Accurate forecasting of local weather patterns is essential for climate resilience and sustainable planning. This study analysed 35 years (1990--2025) of hourly temperature and precipitation data from Thohoyandou, South Africa, to assess the effects of climate change and improve anomaly prediction. Exploratory analysis and BEAST decomposition revealed accelerated warming trends of 0.025 °C per year in temperature anomalies, and irregular rainfall patterns dominated by extreme events rather than systematic changes. Two machine learning models, Artificial Neural Networks (ANN) and Long Short-Term Memory (LSTM) networks, were evaluated for anomaly forecasting, with feature selection guided by LASSO regression. For temperature, the LSTM model outperformed the ANN, achieving RMSE = 0.678, MAE = 0.466, and MASE = 0.520, compared to RMSE = 0.738, MAE = 0.524, and MASE = 0.585 for the ANN, with improvements confirmed by the Diebold--Mariano test. For precipitation, both models performed similarly, with the LSTM slightly better (RMSE = 0.432, MAE = 0.112, MASE = 1.897). These results highlight the LSTM model’s superior ability to capture temperature anomalies and the continued challenges of modelling rainfall, providing evidence-based insights to support agricultural planning, water management, and climate adaptation in Southern Africa.