In continuous casting, it is very important to predict and detect the internal cracks of billet in time for ensuring continuous production, improving product quality and reducing production costs. In this paper, Clustering analysis method is adopted to do feature extraction and classification for on-site data, by which ladder parameter tables of processing parameters and defect grades of internal cracks are got. Fault tree analysis (FTA) method is adopted to analyze the effects of processing parameters on internal cracks. The solidification speed of billet is calculated by solidification heat-transfer model. Quality prediction model of internal cracks in continuous casting billet is established by quality prediction function, based on clustering analysis model of on-site data, FTA model and solidification heat-transfer model. Some samples of Steel Grade 1008 are selected for testing the quality prediction model. The percentage of accuracy for the quality prediction is 80 percent, which provides the foundation for industry application.