Distributed Gaussian Granular Neural Networks Ensemble for Prediction Intervals Construction

Table 5

Statistical analysis of the comparative experiments.

Problems	Model	PICP	MPIW	MAE	PPIM	PPISD	T(s)
Noisy Rossler problem	Conformal prediction in [29]	0.9500	0.5758	0.1742	0.1947	0.0319	0.7623
	MVE-based NN individual	0.8333	0.8917	2.6027	1.4426	0.22	0.5387
	Bayesian NN	0.9500	1.0688	0.3243	1.0688	0.1216	0.7615
	NNs ensemble in [22]	0.9833	1.2238	0.1843	1.2238	0.3881	50.48
	Distributed ESNs ensemble	0.9667	0.6723	0.1506	0.1732	0.0077	57.44
	Distributed NNs ensemble	0.9667	0.5023	0.1356	0.1509	0.0067	1474.14
Noisy Mackey Glass	Conformal prediction in [29]	0.9500	0.4538	0.1149	0.1162	0.0297	0.7623
	MVE-based NN individual	0.9077	0.6482	0.4811	0.3417	0.1921	0.5387
	Bayesian NN	0.9310	0.5084	0.1179	0.1412	0.0385	0.7615
	NNs ensemble in [22]	0.9828	0.5363	0.1178	0.1388	0.0501	50.48
	Distributed ESNs ensemble	0.9527	0.4323	0.1145	0.1211	0.0125	82.42
	Distributed NNs ensemble	0.9722	0.4151	0.1129	0.1116	0.0113	1035.49
Traffic time series	Conformal prediction in [29]	0.9500	12487.00	4135.78	5394.57	295.1779	0.7623
	MVE-based NN individual	0.9962	21674.00	30325.00	17267.00	1282.2291	0.5387
	Bayesian NN	0.8958	12057	4182.77	5789.36	307.6441	0.7615
	NNs ensemble in [22]	0.7615	6525.37	3694.27	4986.57	374.1657	50.48
	Distributed ESNs ensemble	0.7952	4674.72	2705.11	2493.67	168.1368	3.54
	Distributed NNs ensemble	0.7833	3046.309	1752.181	1943.746	129.8434	1108.09
BFG consumption flow	Conformal prediction in [29]	0.9500	7.6245	1.6385	1.7123	0.1265	0.7623
	MVE-based NN individual	0.8077	5.2250	1.9899	2.1357	0.2401	0.5387
	Bayesian NN	0.9167	5.0133	1.7477	1.7046	0.1073	0.7615
	NNs ensemble in [22]	0.8563	4.5125	1.7989	1.9625	0.1758	50.48
	Distributed ESNs ensemble	0.8500	4.3422	1.7285	1.7039	0.0708	14.31
	Distributed NNs ensemble	0.8667	4.2224	1.6232	1.7032	0.0674	828.01