Page 96 - 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

Neuro-Fuzzy Inference System (CANFIS) model has earlier Adaptive Neuro-Fuzzy Inference System
never been used to simultaneously predict hourly, with a multiple-input multiple-output (MIMO)
daily, weekly, monthly and quarterly 3G downlink architecture [21]. CANFIS is an improvement on the
traffic. MISO ANFIS architecture to multiple-input
multiple-output (MIMO) configuration. The CANFIS
CANFIS is a multiple-input multiple-output (MIMO) architecture for five-input five-output is shown in
generalization of the Adaptive Neuro Fuzzy Fig. 1 with five layers. There are five inputs of 3G
Inference System (ANFIS) structure [14]. Many downlink traffic, ݔ ൌ ݄݋ݑݎ݈ݕ ݀ܽݐܽǡ ݔ ൌ
ଶ
ଵ
researchers have explored the advantages of MIMO ݈݀ܽ݅ݕ ݀ܽݐܽǡ ݔ ൌݓ݈݁݁݇ݕ ݀ܽݐܽǡ ݔ ൌ
ଷ
ସ
in the analysis and forecasting in several fields [14] ݓ݈݁݁݇ݕ ݀ܽݐܽ ܽ݊݀ ݔ ൌ ݄݋ݑݎ݈ݕ ݀ܽݐܽ with predicted
ହ
[15] [16] [17]. For instance [17] used CANFIS with hourly, daily, weekly, monthly and quarterly
two inputs and three outputs in fault detection and outputs.
diagnosis of railway track circuits. Reference [17]
applied the CANFIS model to Australian regional The CANFIS structure consists of five layers
flood and concluded that the model provided an whereby each one can be adaptive or fixed in
accurate regional floods estimated level. The performance [22]: Layer 1, Layer 2, Layer 3, layer 4
authors implemented multi-input single output and Layer 5.
(SISO) CANFIS architecture.
Layer 1(Premise parameters): Every node in this
The ability of CANFIS models to work on multiple- layer is a complex-valued membership function ሺߤ )
௜௝
input and multiple-output have been tested by with a node function:
other researchers: 7-input/4-output [18];
9-input/6-output [16]. Reference [19] employed ܱ ଵǡ௝ ൌߤ ሺݔ ሻǡ ݂݋ݎ ݅ ൌ ͳǡ ʹǤ ሺͳሻ
஺௜
ଵ
the CANFIS architecture with 6-inputs and 1-output
to predict farm yields. ܱ ଵǡ௝ ൌߤ ஻௜ିଶ ሺݔ ሻǡ ݂݋ݎ ݅ ൌ ͵ǡ ͶǤ ሺʹሻ
ଶ
Reference [20] evaluated the capabilities of a ܱ ଵǡ௝ ൌߤ ஼௜ିସ ሺݔ ሻǡ ݂݋ݎ ݅ ൌ ͷǡ ͸Ǥ ሺ͵ሻ
ଷ
CANFIS model for the prediction of flow through
trapezoidal and rectangular rockfill dams. The ܱ ଵǡ௝ ൌߤ ஽௜ି଺ ሺݔ ሻǡ ݂݋ݎ ݅ ൌ ͹ǡ ͺǤ ሺͶሻ
ସ
authors in [21] predicted the electric load using the
CANFIS and ANN models and concluded that the ܱ ଵǡ௝ ൌߤ ா௜ି଼ ሺݔ ሻǡ ݂݋ݎ ݅ ൌ ͻǡ ͳͲǤ ሺͷሻ
ହ
CANFIS model outperformed the ANN model.
ZKHUH
The advantage of applying the CANFIS model is that ሺܣ ǡܣ ݋ݎ ܤ ǡܤ ݋ݎ ܥ ǡܥ ݋ݎ ܦ ǡܦ ݋ݎ ܧ ǡܧ ሻ
ଶ
ଶ
ଵ
ଵ
ଵ
ଵ
ଶ
ଶ
ଵ
ଶ
it serves as a single model to predict five different UHSUHVHQWV WKH OLQJXLVWLF YDULDEOH
time spans of telecommunication network traffic, ߤ ሺݔ ሻǡߤ ஻௜ିଶ ሺݔ ሻǡߤ ஼௜ିସ ሺݔ ሻǡߤ ஽௜ି଺ ሺݔ ሻ
ଷ
ଶ
ସ
஺௜
ଵ
unlike the traditional forecasting models, that use
one model for each time span. In previous research ܽ݊݀ ߤ ா௜ are some appropriate parameterized
no study has been conducted that has explored the membership functions (MFs), ݔ ǡݔ ǡݔ ǡݔ ܽ݊݀ ݔ
ଶ
ହ
ଷ
ଵ
ସ
th
forecasting of telecommunication network traffic are the input to the i node.
using the multiple-input multiple-output CANFIS
model with 5-input and 5-output: hourly, daily, Each node in Layer 1 is the membership grade of a
weekly, monthly and quarterly data. fuzzy set ሺܣ ሻ and identifies the degree to which
௜௝
the given input fits to one of the fuzzy sets, which is
2. METHODOLOGY represented in general as equation (6)
The methodology section of this study highlights ܱ ଵǡ௝ ൌหߤ ܣ ሺݖ ሻหנ ߤ ܣ ሺݖ ሻ
௜
௜௝
௜௝
௜௝
௜
௜௝
the approach adopted to instantaneously predict
five-input 3G downlink traffic using the CANFIS ݂݋ݎ ሺͳ൑݅ ൑ ݊ǡ ͳ ൑ ݆൑݉ሻ ሺ͸ሻ
network model and the selection of the best model.

2.1 CANFIS network architecture creation
where ܱ the membership grade of a fuzzy set ܣ ,
௜௝
௜ǡ௝
CANFIS is an extension of the basic principles of the ߤ ௜௝ is any suitable parameterized membership
74 © International Telecommunication Union, 2018