As models trained on the same dataset show different performances, selecting and combining them is becoming more and more valuable. Model selection can improve the error tolerance when considering the models’ performance. It also provides feedback about the strengths and weaknesses of a particular model under a particular condition. Therefore, our motivation is to build a framework that can select the best model under actual conditions. As the input data change, we could be able to get the best results from the model candidates. The data analyzed in this thesis are time-series data, collected from a mechanized tunneling project. The data will be used to build an automated framework for forecasting model selection, and the selected models will be deemed as those most advisable to perform the prediction.