The problem of weather variation and their impacts on socio-economic development is more acute in developing countries like Ethiopia where alternating droughts have been persistent causes of severe economic hardships. Due to this, weather prediction has become a stimulating area of research for scientists and researchers. Weather prediction is a spatio-temporal and time series based process. Predicting future weather condition is a very important issue in today’s world as the precarious fields like air flights, tourism, agricultural and industrial sectors are largely dependent on the weather conditions. Weather prediction involves a combination of computer models, observations, and knowledge of trends and patterns by using which reasonably accurate forecasts can be made up to a finite number of days in advance. This study is aimed to investigate and model the existing weather data series to enable the future information of the weather condition to be forecasted accurately using machine learning techniques. The study presents a research on weather forecasting by using historical weather dataset. Since atmospheric pattern is a complex and non-linear system, traditional methods are seized to be effective and efficient in such situation. It is observed that Artificial Neural Networks, including MLP, GRNN, RBF and Elman recurrent networks, are influential methods for resolving such problems. The criteria used for model selection include MSE, MAE, RMSE, correlation coefficient, and confusion matrix. Among the used predictive models, MLP return acceptable result with the performance criteria MAE and correlation coefficient for temperature and rainfall variables prediction. However, for relative humidity variable prediction MLP model return better result with the performance criteria MSE and correlation coefficient. Therefore, MLP model is sufficient for future weather data prediction for the three selected variables: temperature, rainfall and relative humidity.