<p>Central Ethiopia is a climate-vulnerable area that needs long-term forecasts of temperature variations that would be used to aid in agriculture, water management, and urban development planning. This paper gives an example of a combined statistical and machine learning model of long-term temperature prediction in ten stations in Central Ethiopia on the basis of annual temperature history. The Mann-Kendall test and Sen slope estimator were non-parametric tests applied to determine the degree of significance and magnitude of regional warming by analyzing historical data on maximum (Tmax), minimum (Tmin) and mean (Tmean) temperatures. Linear regression and Elastic Net regression were used to present the main trend elements and Support Vector regression and Gradient Boosting were used to address the remaining variability. RMSE, mean bias error (MBE), MAE and the coefficient of determination (R²) were used as an indicator of model performance on independent test datasets. Findings show significant warming in all the stations that have significant spatial variation. Elastic Net was the best individual predictor as the temperature series is dominated by a trend. More significant improvements in accuracy over the best individual model were not large but a weighted ensemble further enhanced the strength of forecasts by decreasing the variance and extreme deviations. On the whole, the results suggest the significance of matching the complexity of a model to the properties of data when forecasting climatic conditions in data-sparse areas. The suggested framework is a strong and informative one to evaluate the temperature in the long term and offer practical insights into the planning of climate change adaptation and making the decision-making process more risk-resilient in Central Ethiopia and other climatic areas.</p>