Abstract: The semiconductor wafer fabrication system (SWFS) should make changes on the product mix and manufacturing process control frequently to meet customers' fluctuating demands, and the cycle time (CT) forecasting is crucial in the promise of a good delivery-time. A data driven method is proposed to select key factors by estimating the correlation between candidate factors and wafer lots' CT. Firstly, all candidate factors are collected for correlation analysis. Subsequently, a mutual information based method is designed to determine key factors as the input of the forecasting model. Eventually, 78 CT-related factors stood out from 1 202 candidates and replaced 5 global factors (used as reference) to predict CTs of wafer lots. To evaluate the performance, a back propagation network is designed to forecast the CT of wafer lots by using the selected key factors. The results indicate that the proposed approach had higher accuracy than linear regression and FCM-BPN in CT forecasting in large scale dataset.