Table 3 Results for length of stay prediction task (regression). Contrary to mean absolute deviation (MAD), larger kappa is better.

From: Multitask learning and benchmarking with clinical time series data

Model Kappa MAD
Linear regression 0.336 (0.335, 0.338) 116.4 (115.8, 117.0)
Standard LSTM 0.433 (0.432, 0.435) 94.7 (94.2, 95.1)
Standard LSTM + deep supervision 0.413 (0.411, 0.414) 94.5 (94.1, 95.0)
Channel-wise LSTM 0.424 (0.422, 0.425) 94.3 (93.9, 94.8)
Channel-wise LSTM + deep supervision 0.426 (0.424, 0.428) 94.0 (93.6, 94.4)