Research On Evaluation Model Of Innovation Project Based On Improved SVM

Innovation is an important force to promote economic growth in China,and an important part of China's innovation system.In order to carry out the depth cooperation and resource sharing,innovation research at the present time,all over the country has established any of technological innovation platforms to help some other researchers,producers,managers.By extracting and analyzing project data on the technological innovation platforms,it is of great significance to achieve an accurate evaluation of the project.However,technological innovation is an extremely complex and dynamic process which is full of social uncertainly,and it is very huge and extremely complex that project involved the data.In particular,the future cooperation in various fields with the cross-project data on the platform will be more and more.It has become an important problem for managers to provide decision information,when we face how to evaluate the project objectively.To solve above problem,the paper starts at the perspective of data mining and build modeling with machine learning methods.The main work of the paper includes the following aspects:Firstly,it introduces about the background and significance of the technological innovation,and makes an overview of the current status of research scholars about some evaluation questions of collaboration innovation project,and contains solving problems related to the use of theoretical knowledge and analysis of its shortcomings.At last we look into the future for the trends of evaluation model.Secondly,we discuss the theory and techniques involved in the project for building evaluation model,which mainly includes the dimension reduction algorithm,Genetic algorithm,Support Vector Machines algorithm.Thirdly,according to establishing the evaluation index system about project,we get corresponding experimental data with the way of collecting and pre-processing methods.We use SVM classifier to learning and training data with the different kernel functions,and get the optimal kernel function by comparing the prediction accuracy.Later in order to reduce redundant information for data and to improve performance of classifier model to save training time,we specifically apply different dimension reduction algorithms on processing experimental data and input classifier with the post-processing of the sample data for classification tasks.We receive the most appropriate algorithm on dimension reduction aspect and finally built a project evaluation model successfully,which make useof dimension reduction algorithm and combine SVM.Fourthly,after analyzing project evaluation model of LLE+SVM,we pointed out the shortcomings and two related improvements.Mainly starting from the front-end dimensionality reduction and back-end classification,we improved to dimensionality reduction method with supervisory functions and also enhanced effect for data dimension reduction by using category labels with information about the sample data to transform traditional LLE algorithm;for the optimization problem of SVM kernel function parameters and the penalty factor,we exploit improved genetic algorithm about the problem of SVM parameter optimization so as to obtain the best overall performance of SVM.Through the above some related improvements,we established an efficient evaluation model for innovation project.Accurate evaluation is a prerequisite for managing technological innovation projects effectively.it can be efficiently and accurately to build the project evaluation model with paper methods.It has important practical significance for managing and deciding project on technological innovation platforms quickly in the future.