The pattern information detected via data mining is used for predicting in many fields such as pattern recognition, decision making, weather, and transportation. The information and patterns hidden within data have important meaning because they may have a function that has not yet been identified by the existing methods of analysis. However, because most prediction models are still using the algorithm based on supervised learning, they are flawed by various weaknesses such as the issue of preprocessing the training data, difficulties in applying them to new patterns other than the trained data, and difficulties involved in gradual learning of the real time input data. On the other hand, the prediction model that uses only unsupervised learning is flawed by its difficulty in analyzing the result of prediction because no information about the data is given as to when learning is conducted. In order to resolve the weaknesses of both supervised learning and unsupervised learning, this dissertation proposes a hybrid prediction model which integrates the FCM clustering algorithm that belongs to unsupervised learning with the features of supervised learning that lead to collection of target values. The proposed hybrid prediction model conducts automatic classification without external interference, detects target values inside the data alone, and applies them to deriving numerical prediction results. Thus the proposed model possesses the strong features of both supervised learning and unsupervised learning. In this dissertation, We present the experimental prediction results of traffic hours in an intelligent transportation system. We has been confirmed the performance of the proposed Hybrid FCM clustering algorithm through an experiment of travel time prediction for an intelligent transportation system. First, the information required for learning and prediction is extracted from the data sets which are continuously input at t-1 and t points without the user''s data control and preprocess such as class labeling. Secondly, the proposed model has a data-oriented process that performs error correction and prediction through regression analysis at the classified result data sets. We expect that the proposed hybrid prediction model may contribute to enhancement of automation standards in various intelligent systems.