For the random rotation ensemble method,a new dimension reduction algorithm named random transform dimension reduction was proposed,which can reduce the information loss caused by the reduction dimension.After the random transform dimension reduction,the training supervised learning algorithm can obtain higher accuracy and better generalization performance.Through the experiments on the simulated data,it is proved that the regression analysis using multiple collinearity data can retain more information and obtain smaller mean square error than the traditional dimensionality reduction method.The performance of the random transform dimension reduction in handwritten numeral recognition datasets was studied,and it is proved that,compared with the general dimensionality reduction algorithm,the random transform dimension reduction can achieve higher accuracy in image classification.