Semiconductor manufacturing (SM) is one of the most complicated production processes involved deposition, diffusion, ion implementation, lithography, plasma etching and thin film. Bottleneck is the key factor to the SM system which seriously influences the throughput rate, cycle time, time-delivery rate, etc. Efficiently predicting the bottleneck of the system provides the best support for the consequent scheduling. In this paper, Considering that categorical data (the product types, releasing strategies) and numeric data (work in process (WIP), processing time, utilization rate, buffer length, etc) have seriously effect on bottleneck, an Improved Adaptive Network-based Fuzzy Inference System (ANFIS) was adopted to predict bottleneck since conventional neural network-based methods accommodate only numeric inputs. In this improved ANFIS, the contribution of categorical inputs to firing strength is reflected through a transform matrix. In order to tackle high-dimensional inputs, reduce the number of fuzzy rules and obtain high prediction accuracy, a fuzzy c-means (FCM) method combining binary tree linear division method (LD) was applied to identify the initial structure of fuzzy inference system. According to the experiment results, the main-bottleneck and sub-bottleneck of SM system can be predicted accurately with proposed method.