TOPSIS-Sort-C represents each class by a characteristic profile and assigns each alternative to the class with the nearest closeness coefficient. When profiles are built from data, some criterion values may run counter to the class order; the original application addressed such reversals by re-sorting those values across classes. This paper analyzes whether that correction is necessary and how it affects sorting performance. A complete ordering of individual criterion values is shown to be sufficient but not necessary for preserving the class order of profile closeness coefficients, which is the condition the assignment rule actually uses. Re-sorting is further characterized as a dispersion-increasing operation: it never lowers the top profile's closeness coefficient and never raises the bottom one's, and the weighted-sum class representatives after re-sorting majorize those before re-sorting, so the outermost decision boundaries move outward. Because the closeness coefficient is not a monotone transformation of the weighted sum, this boundary result is stated for weighted-sum aggregation. When criterion-level boundaries can be specified, TOPSIS-Sort-B gives the normatively appropriate assignment; since such boundaries are rarely available in practice, that assignment serves as the recovery target for profiles built from class labels alone. On the Index of Economic Freedom data, profiles without re-sorting recover this target more closely while preserving the profile-closeness order, and resampling experiments on the same data give the same result at every profile sample size. Among four other public datasets, re-sorting performs worse on one and ties on the other three, where the reversal is either confined to low-weight criteria or absent; it improves recovery on none of them. A synthetic controlled experiment with known true classes shows that re-sorting reduces accuracy when the reversal reflects the underlying structure, and helps only marginally when a mild reversal arises from noise in very small samples. Criterion values should therefore not be re-sorted mechanically when the profile closeness coefficients already preserve the class order; when they do not, profile construction and the source of the reversal should be reconsidered before any order-restoring correction.

1. 서 론
2. 관련 연구 및 배경
    2.1 관련 연구
    2.2 TOPSIS-Sort-C
    2.3 TOPSIS-Sort-C에서의 재정렬
3. 문제 정의와 연구문제
    3.1 문제 정의
    3.2 연구문제
4. 기준값 재정렬의 필요성 분석
5. 기준값 재정렬의 성능 영향 분석
    5.1 경제자유지수 자료 분석
    5.2 시뮬레이션
6. 결론
References
<부록 A> 합성 통제 실험의 강건성

• Seunghyun Park(Department of Industrial Engineering, Hannam University) | 박승현 (한남대학교 산업공학과) Corresponding author
• Bongwan Choi(Department of Industrial Engineering, Hannam University) | 최봉완 (한남대학교 산업공학과)
• Jihoon Kyung(Department of Industrial Engineering, Hannam University) | 경지훈 (한남대학교 산업공학과)
• Kyeongtaek Kim(Link Nine System, Inc.) | 김경택 ((주) 링크나인시스템)