Experimental data from the three tank system is utilized.For this case study, the problem of predicting the water level in tank 3 (x3) is considered, which shows the characteristics of a low-dynamic signal. From the schematic of the system, it is easy to see that the explanatory variablesfor x3 are the water level in tank 2 (x2) and the flow rate (x4). Models with different number of states are fitted on the training data, and the values obtained for AIC and BIC are plotted against the number of states, as presented in Figure 3a. From this figure, it can be seen that relatively smaller values of AIC and BIC are obtained for models of order 10, 15, and 20. For this case study, we selected a model of 20 states which gives a higher likelihood value (goodness of fit). For the case of IOHMM output functions, Gaussian distributions are fitted, and the meansand variances of the distributions are obtained through a linear regression process between the explanatory variables (x4, x2) and x3.