Construction of a Classification/Decision Tree
0
조회
0
다운로드
3
인용
0
참고문헌
초록
최근 보건 및 의학 등의 전문분야에서 증거-중심 실무 (Evidence-based practice) 를 강조함에 따라 현장 전문가 또는 연구자들이 최선의 의사결정을 위해 다양한 종류의 알고리즘에 관심을 갖고 있다. 증거-중심 실무란 현장 전문가의 임상적 경험과 연구결과의 근거를 통합하여 최선의 의사결정을 이끌어 내는 과정을 의미한다. 현장전문가의 의견, 기존의 지식 및 통계적 방법 등이 임상 전문가들의 의사결정을 이끌어 내기위한 알고리즘의 근거가 되고 있다. 통계적 방법을 적용하는 전통적인 분류법은 크게 선형판별분석과 로지스틱 회귀분석으로 구분할 수 있다. 그러나 이러한 방법들은 현장전문가들에게 복잡한 계산을 요구하는 등의 실용성 측면에서 문제가 있을 수 있다. 이에 반해, 의사결정 나무(decision trees)는 전통적인 방법들과는 달리 복잡한 계산절차가 필요하지 않기 때문에 새롭게 조명받고 있는 방법이다. 이 연구는 의사결정 나무의 두가지 일반적인 방법들 (CART 와 CHAID) 를 소개하고 임상현장에서 이 방법들에 대한 통계적 적용 가능성을 고찰하는 것을 목적으로 하고 있다. 더불어, 의사결정 나무에서 변수를 선택하는 기준과 자료를 정제하는 과정에 대하여 구체적으로 논의하고자 한다.
Abstract
As many professions move towards evidence-based practice the need for established well researched tools increases. In healthcare and medicine often times tools are developed to help identify when a test is neededsuch as radiographs on an acute ankle sprain or a disease is present. These types of demands have produced various types of algorithms, which help clinicians make decisions. Some are based on content experts’opinions while others use known cases and statistics. Some traditional classification methods are linear discriminant analysis and logistic regression. These methods while valid are not usually very practical because it requires clinicians to perform calculations. Other methods such a classification/decision trees have been developed using statistics, but are more user friendly and perform as well and usually better than traditional methods. Classification and Regression Trees (CART) and Chi-Squared Automatic Interaction Detection (CHAID) are twopopular methods for constructing decision trees. This paper will review the background of classification/ decision trees and the methods used to construct them. Issues including splitting and stopping rules along with weights and costs associated with the outcome are addressed.
참고문헌 (15)
- ". Rule induction for group decisions with statistical data - an example. Journal of the Operational Research Society", pp.227~234 (1991)
- ". Predicting stroke inpatient rehabilitation outcome using a classification tree approach. Archives of Physical Medicine and Rehabilitation", pp.619~ (1994)
- ". The use of multiple measurements in taxonomic problems. Annals of Eugenics", 7, pp.179~188 (1936)
- "RainForest A framework for fast decision tree construction of large datasets" (1998)
- "Discovery of predictive models in epidemiologic surveillance databases", pp.359~363 (2000)
- ". An exploratory technique for investigating large quartiles of categorical data. Applied Statistics", 29, pp.119~127 (1980)
- ". Decision trees based on automatic learning and their use in cardiology. Journal of Medical Systems", pp.201~ (1994)
- ". Classification methods for computerized interpretation of electrocardiogram. Methods of Information", pp.330~336 (1990)
- "Split selection methods for classification trees.", 7, pp.815~840 (1997)
- ". A comparison of logistic regression to decision tree induction in medical domain. Comparative and Biomedical Research", pp.74~97 (1993)
- ". The construction of computerized classification systems using machine learning algorithms An overview. Comp in Human Behavior", pp.155~167 (1992)
- "a sequential search program for the analysis of nominal scale dependent variables. Institute for Social Research University of Michigan" (1973)
- ". Problems in the analysis of survey data and a proposal. Journal of the American Statistics Association", 58 (1963)
- "CART: Tree-Structured Nonparametric Data Analysis" (1995), Salford Systems
- ". Comparative evaluation of pattern recognition techniques for detection of misclassifications in mammography. International Journal of Pattern Recognition and Artificial Intelligence", pp.1417~1436 (1993)