Page 100 - ITU Journal - ICT Discoveries - Volume 1, No. 2, December 2018 - Second special issue on Data for Good

ITU JOURNAL: ICT Discoveries, Vol. 1(2), December 2018

Statistic 957 samples 707 samples 178 samples 72 samples 15 samples
(Hourly) (Daily) (Weekly) (Monthly) (Quarterly)
Probability 0.000000 0.000000 0.065425 0.010755 0.323506

Table 4 – CANFIS architecture training results

CANFIS Architecture MSE NMSE R % Error
(Training: Validation: Testing)
(70%:15%:15%) 0.00298 0.0303 0.71 32.34
(70%:10%:20%) 0.00373 0.0339 0.70 34.89
(80%:10%:10%) 0.00460 0.0377 0.698 36.70
(60%:10%:30%) 0.0034 0.0328 0.683 36.24

Table 5 – Selection of network architecture for forecasting
CANFIS Model 5-3-5 5-3-5 5-2-5 5-2-5 5-2-5

Transfer function TanhAxon TanhAxon TanhAxon Axon TanhAxon
Learning Rule Step RProp RProp Momentum Step
MSE 0.0092 0.0021 0.00108 0.000406 0.00559
NRMSE 0.0531 0.0252 0.01825 0.01120 0.0415
R 0.5968 0.7868 0.9268 0.9737 0.6262
% Error 48.37 21.05 17.82 12.33 52.55

The CANFIS (5-2-5) model testing window with
convergence rate for the data is shown in Fig. 2. The
active cost curve plot indicates that with testing
data the algorithm has successfully undergone
generalization and thereby converging to zero.

3.3 Network Validation

The CANFIS (5-2-5) network for five-input five-
output data with 2 MFs per input, was validated by Fig. 2 – CANFIS (5-2-5) testing window
comparing the actual and predicted traffic as The study found out that with the increase in the
demonstrated in Fig. 3 for hourly traffic prediction. number of inputs, the CANFIS model produce
accurate traffic forecasting when membership
The daily, weekly, monthly and quarterly prediction functions per input are 2. Therefore the best model
of 3G traffic are exhibited in Fig. 4, Fig. 5, Fig. 6 and for predicting five-input telecommunication traffic
Fig. 7. The CANFIS model employed the Bell was CANFIS (5-2-5).
membership function, Axon transfer function and
Momentum learning rule. The membership function
per input was varied between 2 and 7.

78 © International Telecommunication Union, 2018