The Classification Of Symbol-valued Data Based On Random Weight Networks

With the coming of big data era,the scale of data is becoming larger and larger,and at the same time,the data types are various,which may be numeric type,symbolic value data or mixed type.It has become a hot research topic in the field of machine learning and has important applications that how to effectively mine the knowledge from various massive data,including symbolic value data.The classification problem is one of the main research problems in machine learning.The paper focuses on studying the classification of symbol-valued data based on random weight networks(RWNs).RWNs are also called Extreme Learning Machine(ELM).The main idea of RWNs is to speed up the learning by randomization methods.This paper investigates the RWNs with symbolic attributes,and experimentally compared it with C4.5 algorithm on three aspects:(1)the computational time complexity and generalization ability;(2)the impact of sample size on the performance of algorithms;(3)the ability to handle imperfect data of various types.The following valuable conclusions are obtained:(1)ELM has very fast learning speed with competitive testing accuracy.The differences of testing accuracies of C4.5 and ELM are not significant.(2)It is not always true that the testing accuracies increase with the increase of number of samples.However there is no significant difference between the fluctuations of C4.5 and ELM.(3)ELM has almost same ability to handle the imperfect data.