The Visualization And Pattern Recognition Method Based On The Fusion Of Incremental Learning And Feature Extraction

The advent of the information age makes every field of human life produce various and changing data and the scale of data is increasingly massive.According to the human progressive cognitive principle,it is precisely challenging to acquire a complete pattern of knowledge in a relatively extensive data background.Moreover,learning is regarded as the process of classifying new knowledge from the original knowledge pattern.The idea of incremental learning accords with the general law of progressive progress in acquiring knowledge,therefore it is essential to introduce the incremental learning algorithm in machine learning algorithms providing them with the dynamic updating learning ability.The attribute partial order structure has the characteristics of definite hierarchy,simple structure,and none crossover,and it is a data visualization tool to characterize the association relationship between the concepts.At present,the graph algorithm of attribute partial order structure bases on the method of batch construction,in this paper,an incremental learning algorithm was integrated into graph algorithm,which makes it gain the function of generating graph dynamically,and realize the data mining process of the simplest and entire mode.With the realization of data granulation and incremental learning,the data dimension will also be improved increasingly.In order to select high-value features and reduce the impact of noise data on classification,this paper introduces the Lasso(Least absolute shrinkage and selection operator)algorithm to select the attribute feature of qualitative data and quantitative data,not only to ensure the accuracy of pattern recognition but also to benefit the visualization effect.In this paper,the research status of the incremental learning algorithm,granulation,and formal concept analysis are analyzed,and the theory knowledge of mixed data processing,attribute partial order principle and covering principle are introduced.Besides,a continuous data granulation algorithm based on the minimum Gini index and a form background of the mixed data processing method is proposed as the main line of the graph algorithm of attribute partial order structure.To solve the problem of increasing the number of feature dimensions in incremental learning,a method combining Lasso algorithm and feedback model is designed to select the optimal local combination under the precondition of the specific classification rate.At the same time,a new algorithm based on CGAO(Combination of Gini And Objects)is put forward to optimize the data structure,and a computer generating method of layered pattern matrix is also proposed.Finally,the rule visualization is realized combining the attribute partial order structure diagram with the classification method proposed,besides,five databases from UCI and classifiers(1NN,3NN,SVM,Adaboost,Random Forest)are selected to make comparison with the precision results of our method.When the learning proportion reaches 40%,the precision of the Pima Indians Diabetes database(77.66%)can exceed that over the Adaboost(75.32%),SVM(77.27%),1NN(59.74%)and 3NN(75.97%)algorithms.The precision of the Pima Indians Diabetes database reached 80.52% when the learning proportion is 90%.Furthermore,the necessity of visual incremental learning model is also validated.