Benchmarking methodology for multiclass classification models based on multi criteria decision analysis
The purpose of this research was to develop a benchmarking methodology for aidingmedical organizations administrations in benchmarking and ranking availablemulticlass classification models to select the best one. Medical organizations havebeen facing difficulties in evaluating and comparing classifi...
Saved in:
| 主要作者: | |
|---|---|
| 格式: | thesis |
| 语言: | eng |
| 出版: |
2019
|
| 主题: | |
| 在线阅读: | https://ir.upsi.edu.my/detailsg.php?det=6355 |
| 标签: |
添加标签
没有标签, 成为第一个标记此记录!
|
| 总结: | The purpose of this research was to develop a benchmarking methodology for aidingmedical organizations administrations in benchmarking and ranking availablemulticlass classification models to select the best one. Medical organizations havebeen facing difficulties in evaluating and comparing classification models.Experimental and case study research methods were adopted in this study. The newbenchmarking methodology was proposed based on two stages. In first stage, aDecision Matrix (DM) was constructed based on the crossover of two groups ofmulti-evaluation criteria and 22 multiclass classification models. The matrix wasevaluated using secondary datasets consisting of 72 samples of acute leukemia,including 5327 gens. In the second stage, multi-criteria decision-making techniques,namely, Best and Worst method (BWM) and Vlse Kriterijumska OptimizacijaKompromisno Resenje (VIKOR) were used to benchmark and ranked the multiclassclassification models. The BWM was applied to calculate the weights of evaluationcriteria, whereas VIKOR was used to benchmark and rank the multi-classclassification models. VIKOR was utilized in two decision-making contexts, namelyindividual and group contexts. In group decision making, internal and external groupaggregations are applied. For validating the proposed methodology, an objectivemethod was used. The results showed that (1) the integration of BWM and VIKORwas effective for solving the benchmarking/selection problems of multi-classclassification models. (2) The ranks of multi-class classification models obtained frominternal and external VIKOR group decision making were almost the same, where,Bayes. Nave Byes Updateable, Bayes Net, Decision Stump were the first threeclassification models respectively and Trees. LMT was the last one. (3) In theobjective validation, the ranking results of internal and external VIKOR groupdecision making were valid. Clearly, as a conclusion, the proposed methodology canbe used for evaluation and benchmarking different multiclass classification models forvarious applications. The implications of this study will benefit medical organizationsby enabling them to make the right decisions regarding the use of multi-classclassification models for acute leukemia and the implications also benefit medicalclassification software developers who work in industrial companies and institutionsin developing classification models. |
|---|