. 2021 Mar 22;8(21):16035–16046. doi: 10.1109/JIOT.2021.3067605

TABLE V. Normalized Confusion Matrices (in [%]) of the Best Models in Each Task on the Test Set. Task: Late Fusion of Open Smile: ComParE Func. + SVM, :.01; DeepSelf: e2e CNN+RNN; OpenXBOW: ComParE BoAW + SVM, :.01, : 125; auDeep: RNN+SVM, :.1, : −60 dB. Task: Late Fusion of OpenSMILE: ComParE Func. +SVM, :.0001; auDeep: RNN+SVM, :.1, : Fused; DeepSpectrum: VGG19 +SVM, :.01; OpenXBOW: ComParE BoAW +SVM, : 1.0, : 250; DeepSELF: e2e CNN, Channel: [3, 6], Kernel Size: [16, 8], Stride Size: [16, 8], Learning Rate:.0001. A Task: OpenSMILE: ComParE Func. +SVM, :.0001. (a) Task (UAR = 44.3%, Chance Level: 33.3%). (b) Task (UAR = 44.4%, Chance Level: 33.3%). (c) Task (UAR = 55.3%, Chance Level: 33.3%).