[C76] E. Kalapanidas, N. Avouris, M.Craciun and D.Neagu, Machine Learning Algorithms: A study on noise sensitivity, in Y. Manolopoulos, P. Spirakis (ed.), Proc. 1st Balcan Conference in Informatics 2003, pp. 356-365,Thessaloniki, November 2003. (pdf)
In this study, results of a variety of ML algorithms are tested against artificially polluted datasets with noise. Two noise models are tested, each of these studied on a range of noise levels from 0 to 50algorithm, a linear regression algorithm, a decision tree, a M5 algorithm, a decision table classifier, a voting interval scheme as well as a hyper pipes classifier. The study is based on an environmental field of application employing data from two air quality prediction problems, a toxicity classification problem and four artificially produced datasets. The results contain evaluation of classification criteria for every algorithm and noise level for the noise sensitivity study. The results suggest that the best algorithms per problem in terms of showing the lower RMS error are the decision table and the linear regression, for classification and regression problems respectively.