In this study, a forecast of the monthly Construction Cost Index (CCI) for Türkiye was conducted using the Temporal Fusion Transformer (TFT) architecture, one of the advanced deep learning algorithms. The primary aim of the study is to introduce the TFT algorithm, which has recently gained attention as a powerful deep learning approach for forecasting, and to evaluate its performance by applying it to the prediction of the Construction Cost Index. For the CCI forecasting, monthly data covering the period from January 2015 to September 2025 were utilized. In addition, three independent variables that are considered to potentially influence cost dynamics were selected and incorporated into the model: the number of construction buildings (m²), the industrial production index, and the housing sales index. During the forecasting process, the performance of the TFT model was evaluated using the Root Mean Square Error (RMSE) metric. The RMSE value obtained for the TFT model was calculated as 36,01. The results indicate that the TFT model achieves a high level of accuracy in both short-term forecasts and overall period performance. Moreover, the model successfully captures the underlying trend and volatility dynamics of the construction cost index. Furthermore, through variable importance analysis, the relative contributions of the key indicators affecting construction costs were quantitatively identified, providing additional insights into the factors influencing construction cost dynamics.