Research On Machine Learning Algorithm Based On Expert Knowledge

Machine learning is a discipline that uses computers to simulate human learning behavior.Computer-aided diagnosis and economic data prediction are important application areas for machine learning.For the above two practical applications,we designed two machine learning prediction algorithms based on expert knowledge and achieved excellent prediction performance.In addition,in this paper,we also designed a multi-task regression prediction algorithm,which used information from related tasks to improve the accuracy of the target task's prediction."Total retail sales of consumer goods" is an important indicator of national economy.Based on the functional characteristics of data,we used functional data analysis methods to predict the total retail sales of consumer goods in China.The total amount of data is decomposed into long-term trend components and seasonal fluctuation components.For the prediction of long-term trend components,we introduced the household disposable income index to assist in forecasting.For seasonal components adopted weighted prediction method of the adaptive weighted selection,then we used the sum of the two forecasted values as the total forecasting result.Experimental results on real data showed that our proposed prediction method has a small prediction error,and the prediction results are very interpretable.Glaucoma is one of the most frequent causes of blindness,and early diagnosis is difficult.In this paper,we proposed a glaucoma classification algorithm based on expert knowledge.We used the idea of principal component analysis(PCA)to design a priori PCA algorithm by using the vertical cup plate ratio(VCDR)and ISNT score as a prior direction.Then we extracted the multi-dimensional features of the fundus based on the prior PCA algorithm,and used the classification algorithm in machine learning to predict the two categories.In 478 cases of data we obtained a prediction accuracy of 97.71%.Multitask learning is an inductive migration mechanism,which is intended to improve the generalization ability of the model by using implicit information of multiple related tasks.In this paper,we designed a multi-task least squares regularization regression algorithm,which using relevant tasks of the training sample data to choose the most suitable model parameters for target task,and then uses the sample data of the target task to learn the target task's regression function.Compared with other classical multi-task algorithms,the prediction results are preferable to those of other classical multi-task algorithms.