In precision metal machining, internal heat is generated within CNC lathes as components such as the spindle, turret, and hydraulic system operate. This heat is transmitted to the machine structure, leading to thermal deformation, which can significantly contribute to overall machining errors. Additionally, ambient temperature fluctuations further impact the geometric stability and machining accuracy of the machine tool. To address this issue, this study develops a thermal deformation prediction model based on a Gated Recurrent Unit (GRU) neural network, aiming to accurately forecast spindle displacement induced by thermal effects. The GRU model is well-suited for capturing nonlinear and time-dependent patterns in sequential data, enabling dynamic and reliable predictions. For data acquisition, two displacement sensors and multiple temperature sensors are installed at key locations to monitor thermal-induced displacements. During preprocessing, the Least Absolute Shrinkage and Selection Operator (LASSO) regression is applied to identify temperature variables most sensitive to thermal deformation. A heatmap is subsequently used to visualize the high collinearity among these sensors, providing insights into sensor interactions. Experimental results indicate that the proposed GRU-based model yields accurate predictions of spindle displacement. Model performance is evaluated using four metrics: the coefficient of determination (R-square), root mean square error (RMSE), worst-case prediction error (WErr), and prediction robustness (Rbst). The findings demonstrate that the GRU model delivers high predictive accuracy and stability, making it a promising approach for thermal deformation prediction in CNC machining.