Abstract: Stock data has the characteristics of nonlinearity and lots of noise.Traditional stock prediction models cannot sufficiently recognize the nonlinear features and reduce noise in stock data,which leads to low prediction accuracy.In order to improve the prediction accuracy,remove redundant attributes and enhance discriminability of features,this study cites linear local tangent space alignment from manifold learning and proposes a novel stock price prediction model based on optimized support vector regression.Firstly,linear local tangent space alignment algorithm is used to extract features of original stock data.Then support vector regression is utilized to model the nonlinear relationship between those features and stock close price.Meantime genetic algorithm is used to optimize parameters of support vector regression.Finally,the prediction accuracy is improved.To validate the effectiveness of the model,S&P 500 index stock data from the year 2012 to 2013,and the year 2014 to 2015 is adopted.The experiments prove that the proposed model outperforms others and has higher accuracy and more powerful generalization ability.