In this paper non-invasive disease prediction models are created for stroke, digestive disease and respiratory disease. The invasive disease prediction which causes wounds to human bodies is unsatisfactory. This is the primary motivation for our research. In our experimentation design three kinds of factors including meridian indices, past disease records and daily living habits are supposed to be potential indicators for determination of non-invasive disease forecast. The database provided by Hong Kong Intelligent Health Management Limited Corporation records 18560 samples from the year 2006 to 2010. We employ the popular data mining technology of support vector machine in building prediction models. We succeed in finding high-quality prediction model for digestive disease. The extrapolated prediction accuracy achieves more than 95%. We also find robust prediction models for stroke and respiratory disease. Another contribution of this paper results from the proposition about the assembled discrimination method. We prove that increase in independent discrimination functions will improve the prediction results. It can be used in statistical inference, discrimination analysis and pattern recognition and so on.