Artificial Neural Networks (ANNs) play a significant role in emulating Building Energy Simulation (BES), forecasting building energy consumption, and optimizing energy retrofit measures. The determination of the appropriate ANN's architecture is a complex issue. Hyperparameter optimization (HPO) involves aligning ANN prediction results with data to achieve optimal performance by tuning the ANN's hyperparameters. This method is applied to retrofit an existing low-energy building. To construct the ANN for retrofit measures, data obtained through Latin Hypercube sampling are utilized. The ANN is used to predict the hourly energy consumption, the hourly energy generation, and thermal comfort in the retrofit scenario. The results underscore the importance of HPO in achieving predictions with an ANN, demonstrating an R² error exceeding 0.9.