摘要 |
<p><P>PROBLEM TO BE SOLVED: To predict probability of the state that should be predicted accurately considering interaction and nonlinear contribution of each explanation variable. <P>SOLUTION: The prediction method of this invention is provided with a read-out step which reads out from database study data that is a set of samples having a plurality of explanatory variables and a dependent variable that shows presence of occurrence of a certain state, a coefficient calculation step which seeks for a regression coefficient at a defined regression using a kernel function as the sum of element kernel functions prepared for every above explanatory variables by optimizing objective function given beforehand using the learning data, a dependent variable calculation step which seeks for dependent variables by inputting a plurality of explanation variables as input parameters to the regression, and a probability prediction step which predicts a probability by which the certain state occurs or does not occur inputting the sought dependent variable to a probability prediction function. <P>COPYRIGHT: (C)2007,JPO&INPIT</p> |