@article{oai:doshisha.repo.nii.ac.jp:00028597,
 author = {陳, 彦如 and Chen, Yanru and 鄭, 弯弯 and Zheng, Wanwan and Jin, Mingzhe and 金, 明哲},
 issue = {3},
 journal = {同志社大学ハリス理化学研究報告, The Harris science review of Doshisha University},
 month = {Oct},
 note = {最大情報系数係数(MIC)は非線形相関関係も計測可能な相関係数として2011年に提案され、多くの人工データにおける有効性が示されている。しかし、実データにおけるその有効性に関する実証は十分行われていない。本研究は、様々な分野から集めた多種なベンチマークデータ30セットを用いて、分類問題の側面からMICの有効性を検証した。用いた分類手法は距離判別法とサポートベクトルマシンであり、評価は正解率と計算コストに焦点を当てた。その結果、ベースラインのユークリッド距離、ピアソン相関係数、スピアマン相関係数より、MICを用いた場合の分類精度は劣り、計算コストははるかに高いことがわかった。, The Maximal Information Coefficient is a measure that proposed in 2011 and can detect non-linear relationships in artificial experiments. However, its effectiveness of real-world data is not sufficiently proved. In this study, various of benchmark datasets from different fields were gathered to evaluate the effectiveness of the Maximal Information Coefficient in real-world classification cases. The distance-based discrimination and support vector machine were adopted as classifiers, while accuracies and computational costs are employed to evaluate the results. According to the results, comparing with the baselines including Euclidean distance, Pearson correlation coefficient (Cosine similarity) and Spearman's rank correlation coefficient, the classification accuracy of Maximal Information Coefficient fails to show its superiority, and its computational costs are significantly higher than other measures., application/pdf},
 pages = {149--156},
 title = {The effectiveness of maximal information coefficient in real-world classification tasks},
 volume = {62},
 year = {2021},
 yomi = {チン, ゲンジョ and テイ, ワンワン and キン, メイテツ}
}