Graphical Programming And Application Of Data Prediction

Data prediction has demand in many aspects,but it is difficult to use it in practice.Various prediction algorithms have been proposed at present,and the scope of applicability of these algorithms is different,which makes it difficult for programmers to use it.It requires professional knowledge to carry out the model design of data prediction,and it also needs to have some knowledge of applied programming in order to make it successful.In order to solve these problem,the idea of graphical programming is applied to the data prediction based on machine learning,discusses how to simplify the data prediction algorithm based on machine learning from both theoretical and practical aspects,and finally realizes a data prediction platform that can also be used by non-professional machine learning users,and some steps are studied and improved.The research and innovation points of this paper are as follows:(1)Currently there is no clear distinction between the various algorithms in machine learning,especially in the application of machine learning to data prediction algorithms,which makes it impossible to sort algorithms in order of execution.Therefore,it is impossible to determine whether there is any possibility that algorithms can be used together.In order to solve this problem,a partitioning method of machine learning is proposed by referring to the knowledge in the big data,data science and machine learning.In this paper,machine learning is divided into three major steps(feature engineering,model design and result evaluation),and a total of ten small steps.Through this partitioning method the algorithm can be distinguished in order,then the algorithm of each layer is combined with the background of data prediction,the visual programming will be achieved.(2)In the context of practical application of research,the source of the data can be unreliable.The method of reliable evaluation is needed.The sliding windows mechanism for reliable evaluation usually requires traversing the data in the entire window,when algorithms like KNN and SVM are used,The time complexity of these algorithms is not less than O(n2),and data outside the sliding window is not used.To solve these problems,the generative adversarial network is applied to reliable evaluation of data.The generative adversarial network can consider all the traversed data,and the time complexity relative to the data size is O(n).(3)The time complexity of feature selection algorithm based on soft set theory is exponential.To improve the execution speed of the algorithm,combining soft set theory with probability theory,the concept of soft variance is proposed.The significance of soft variance as an index of soft set feature selection is verified theoretically,an approximate algorithm for soft set parameter reduction is proposed.Compared with the traditional algorithm with exponential time complexity and 01 linear programming algorithm with uncertain complexity(usually regarded as exponential complexity),the time complexity of the proposed algorithm is O(n)in the context of big data.Through the above research,using the current Web technology,the data prediction platform that can be applied by non-professional machine learning users is realized.Considering that machine learning algorithms consume a lot of computing resources and the project is complex,Spring Cloud is adopted as the framework of server design in this paper,In order to realize the graphic program design,we use the current popular Web end framework VUE +Element UI,the Python language is used to design machine learning algorithms.In order to realize multilingual support in the future,the Python computation is serviced by using remote call technology.In order to simplify the usage conditions,a resource-oriented data prediction system is constructed using the design idea of the REST API.Through the system,a comprehensive test experiment is carried out on the commonly used data prediction methods,and five steps are selected from the feature engineering and prediction algorithms.The influence of these five steps on the accuracy of the final data prediction algorithm is tested respectively.The experiment shows the practical effect of each stage in machine learning.