Research On Prediction And Evaluation Methods Based On Enterprise Data

With the development of science and technology, especially advances and more and more applications have been made in the Internet, computer and communication area, it is inevitably that people have been overwhelmed by the ocean of data. Whether it is in the daily life or the enterprises’ production and operation activities, data have been emerging in anytime and anywhere. Since the data generating channels increased unceasingly, it makes the ways of people collecting data much easier and the costs are correspondingly reduced constantly. But, in the meantime, the increasing amount and more and more complicated forms of the data have also led to more difficult exploitation in this area. So the problems such as how to process and exploit the data effectively to make full use of the data and provide some supportive methods for the human life and social development have became the topics in all circles of the society. Due to the data issues are constantly heating up, the global academia, industrial sectors and even governments are highly concerned about the related data techniques. Consequently, data science has been becoming a new scientific research area gradually. On the account of rational use of the data not only can serve the public well, but also can bring enormous commercial value for the enterprises, the data had been referred to "the oil in the future". Meanwhile, the competition on the possession, controlling and exploitation of the data will become more and more fiercely. Although the importance of the data has been recognized by the whole world, many domestic enterprises still unable to enjoy the benefits brought by data for the imbalance of social development. Many enterprises are still in the initial stage of the data exploitation and their precious remains waiting for wake up.Based on widely collection and comprehensive analysis of various studies on data processing and exploitation in the domestic and foreign enterprises and field investigation carried on corresponding enterprises, the dissertation made the determination of the research field by combining the problems existed in the course of data processing and exploitation in the corresponding enterprises recently. Firstly, on account of complexity and dimensionality of data in the enterprises have been increasing ceaselessly and difficulties of data application in enterprises, the dissertation have studied on the related data preprocessing methods. Secondly, according to cost controlling and customer relationship management issues in the enterprises, the prediction methods have been studied in this dissertation based on field investigation and data analysis and processing. Thirdly, in view of improving the effectiveness and scientificity of the enterprise management, the evaluation method has been studied in this dissertation.The specific research contents and innovative work of this dissertation are as follows:1 In order to effectively exploitation of enterprises’ data, the dissertation have studied on the related data preprocessing methods based on characteristics of enterprise data. Firstly, according to idea of maximize the difference between the categories, the dissertation have constructed a corresponding dimensionality reduction method. Secondly, a modified linear dimensionality reduction method also has been studied in this dissertation. The related experiments showed that the methods have played certain positive roles in the effective exploitation of enterprise data.2 Based on intensive studies on the principles of deep learning models, the dissertation has proposed two types of hybrid deep learning structure. Meanwhile, in order to verify the validity of the models, the dissertation have collected some data tables from a telecommunication enterprise. Then the dissertation have extracted a corresponding dataset by sorting the tables in Oracle database. By corresponding experiments, the sequential-hybrid deep learning model showed a better result.3 The dissertation has designed an improved prediction model based on the concept of using the classification errors of CART classifier to adjust the relative weights of Boosting algorithm automatically. Meanwhile, we applied the model to a telecommunication enterprise for customer churn prediction. In addition, the dissertation also have utilized the histogram test and Chi-squared test to determine which attribute should be used in the model. The corresponding simulation experiments showed that the model designed in this dissertation have better results.4 By combining the renewal process and renewal theory, the dissertation have described the maintenance process essentially in a mathematical way. Then we proposed a method for predicting maintenance replacement rate. By applying the method to a MRO enterprise, the replacement probability distribution of a spare part have been fitted out by transformation of the corresponding data. Finally, the parameters of Weibull distribution have been estimated by using the least squared method. By integrating the replacement rate of the spare parts with the BOM information, MRO enterprises can control its costs and provide supportive assistance for the lean MRO.5 Based on analogy of the evaluation issues with the state-changing problem in the thermodynamic system, the dissertation has proposed an evaluation method by caculating state entropy of the evaluation target. Meanwhile, the dissertation also have presented the corresponding mathematical deduction of the state entropy formula. When applying the method to a MRO enterprise, the dissertation have built a corresponding evaluation index system by BSC on the viewpoint of the enterprise. The the weights of indexes have been processed by ANP method accordingly. The model in this dissertation showed a certain effectiveness by the corresponding experiment.