임상자료를 이용한 나무구조 분류모형의 성능 비교
Comparison of Various Classification Tree Methods with Clinical Data
0
조회
0
다운로드
7
인용
0
참고문헌
키워드
초록
Objectives: A classification tree is one of the statistical tools that is widely used in the data mining field. It is useful for making statistical decisions, for example, in medical, biology, and business management area. In this paper, we examine newly developed classification tree algorithms and compare them with real examples in medical study, and provide a guideline to select appropriate methods for data analysis. Methods: For the comparison, we used four clinical datasets from UCI (University of California, Irvine) repository. We divide each data to 2/3 training and 1/3 test data set. After fitting the models with various R packages (tree, rpart, party, evtree, CORElearn and randomForest), misclassification rates for training data and test data are calculated separately. Also, specificity and sensitivity are calculated for test data. This procedure is repeated 200 times and compare misclassification rates with one-way analysis of variance and Tukey’s honest significant difference (HSD). Also, specificities and sensitivities are compared. Results: In every case, randomForest shows the best performance. For the single tree methods, the performance of methods is different in each data set. evtree show better performance than the other methods in most data sets. Most sensitivities in Breast Tissue and Dermatology data are quite large. rpart and ctree show very low specificity in Dermatology Data. Conclusions: Every method has its own characteristic and the performance depends on data. Our study shows that the best single tree methods are different in four example data and evtree shows slightly better performance than the other single tree methods in most data sets. randomForest always shows the best performance, mainly because of using a lot of trees instead of one tree.
Abstract
Objectives: A classification tree is one of the statistical tools that is widely used in the data mining field. It is useful for making statistical decisions, for example, in medical, biology, and business management area. In this paper, we examine newly developed classification tree algorithms and compare them with real examples in medical study, and provide a guideline to select appropriate methods for data analysis. Methods: For the comparison, we used four clinical datasets from UCI (University of California, Irvine) repository. We divide each data to 2/3 training and 1/3 test data set. After fitting the models with various R packages (tree, rpart, party, evtree, CORElearn and randomForest), misclassification rates for training data and test data are calculated separately. Also, specificity and sensitivity are calculated for test data. This procedure is repeated 200 times and compare misclassification rates with one-way analysis of variance and Tukey’s honest significant difference (HSD). Also, specificities and sensitivities are compared. Results: In every case, randomForest shows the best performance. For the single tree methods, the performance of methods is different in each data set. evtree show better performance than the other methods in most data sets. Most sensitivities in Breast Tissue and Dermatology data are quite large. rpart and ctree show very low specificity in Dermatology Data. Conclusions: Every method has its own characteristic and the performance depends on data. Our study shows that the best single tree methods are different in four example data and evtree shows slightly better performance than the other single tree methods in most data sets. randomForest always shows the best performance, mainly because of using a lot of trees instead of one tree.
참고문헌 (16)
- Morgan JN, "Problems in the analysis of survey data, and a proposal", J Am Stat Assoc, 58(302), pp.415~434 (1963)
- Messenger R, "A modal search technique for predictive nominal scale multivariate analysis", J Am Stat Assoc, 67(340), pp.768~772 (1972)
- Quinlan JR, "Induction of decision trees", Machine learning, 1(1), pp.81~106 (1986)
- Breiman L, "Classification and regression trees" (1984), CRC press
- Kass GV, "An exploratory technique for investigating large quantities of categorical data", Applied Stat, 29(2), pp.119~127 (1980)
- Loh WY, "Tree-structured classification via generalized discriminant analysis", J Am Stat Assoc, 83(403), pp.715~725 (1988)
- Quinlan JR, "C4. 5: programs for machine learning" (1993), Elsevier
- Loh WY, "Split selection methods for classification trees", Stat Sinica, 7, pp.815~840 (1997)
- Kim H, "Classification trees with unbiased multiway splits", J Am Stat Assoc, 96, pp.598~604 (2001)
- Hothorn T, "Unbiased recursive partitioning: a conditional inference framework", J Compu Graph Stat, 15(3), pp.651~674 (2006)
- evtree: evolutionary learning of globally optimal classification and regression trees in R, "Grubinger T", Working Papers in Economics and Statistics, ~, pp.No. 2011-20 (2011)
- Robnik S, "CORE-a system that predicts continuous variables", Proceedings of Electrotechnical and Computer Science Conference (ERK’97) (1997)
- "An introduction to recursive partitioning using the RPART routines"[URL]
- Robnik SM, "An adaptation of Relief for attribute estimation in regression", pp.296~304 (1997), Morgan Kaufmann Publishers Inc
- Li M, "An introduction to Kolmogorov complexity and its applications" (2013), Springer Science & Business Media
- Breiman L, "Random forests", Machine Learn, 45(1), pp.5~32 (2001)