TLGRU: time and location gated recurrent unit for multivariate time series imputation

EURASIP Journal on Advances in Signal Processing

Table 1 The MSE results of the proposed method and other imputation methods on the KDD dataset. In most cases, our method owns the best imputation accuracy

Missing rate	Median	Mean	KNN	MF	ISVD	GAIN	\({E}^{2}\) GAN	Ours
10%	0.519	0.386	0.344	0.361	0.356	0.359	0.334	\({\alpha }_{t}\)=1,\({\alpha }_{l}\)=0.5 0.336
20%	0.627	0.522	0.538	0.526	0.544	0.523	0.517	\({\alpha }_{t}\)=0.8,\({\alpha }_{l}\)=0.3 0.510
30%	0.683	0.608	0.628	0.621	0.620	0.616	0.584	\({\alpha }_{t}\)=0.3,\({\alpha }_{l}\)=0.3 0.578
40%	0.755	0.659	0.682	0.681	0.708	0.663	0.641	\({\alpha }_{t}\)=1,\({\alpha }_{l}\)=0.2 0.637
50%	0.816	0.735	0.733	0.760	0.740	0.714	0.694	\({\alpha }_{t}\)=0.5,\({\alpha }_{l}\)=0.3 0.697
60%	0.821	0.736	0.734	0.727	0.714	0.724	0.705	\({\alpha }_{t}\)=0.4,\({\alpha }_{l}\)=0.2 0.702
70%	0.855	0.776	0.807	0.766	0.759	0.768	0.745	\({\alpha }_{t}\)=0.6,\({\alpha }_{l}\)=0.2 0.741
80%	0.883	0.816	0.831	0.792	0.824	0.787	0.756	\({\alpha }_{t}\)=1,\({\alpha }_{l}\)=0.8 0.747