Page 97 - 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

function and ݖ is a real number. ݂݋ݎ ሺͳ ൑ ݅ ൑ ݉ሻ ሺ͹ሻ
௜
Layer 2 (Firing strength): Every node in this layer is ܱ ଶǡ௝ ൌߤ ሺݔ ሻߤ ሺݔ ሻߤ ሺݔ ሻߤ ሺݔ ሻߤ ሺݔ ሻǡ
ଷ
ସ
஽௜
஼௜
ଵ
஺௜
ଶ
஻௜
ହ
ா௜
the product of all the incoming signals. This layer ݂݋ݎ ൌ ͳǡʹǡ͵ǡͶǡͷǤ ሺͺሻ
receives input in the form of all the output pairs
from the first layer: where ݓ is the weights equivalent to the jth MF of
௜௝
input i.

ܱ ଶǡ௝ ൌݓ ൌߤܣ ሺݖ ሻߤܣ ሺݖ ሻǡǥ ǡ ߤܣ ௜ଵ௡ ሺݖ ሻ
௜ଵ
௜ଶ
௡
௜
௜௝
ଶ

Fig. 1 – 5-input 5-output CANFIS architecture for telecommunication traffic

parameters, n is the number of rules, j is the
Layer 3 (Normalised firing strength): Every node in number of outputs.
this layer calculates rational firing strength using
the formula: Layer 5 (Overall output): This layer computes the
output of the CANFIS network using equation (11)
ܱ ଷǡ௝ ൌݓതതതത ൌ σ ೘ ௪ ೕ ௪ ೕ ݂݋ݎ ሺͳ൑݆൑݉ሻ ሺͻሻ [20]:
ఫ
ೕస೔
ܱ ହ ǡ௝ ൌ෍ ݓതതതതߤ ሺͳͳሻ
ఫ
௝
where ݓതതത is the output of layer 3.
ఫ
where ܱ is the overall output, j= 1,2,3,4,5.
Layer 4 (Consequence parameters): Every node in ହ ǡ௝
this layer is a multiplication of normalized firing
strength from the third layer and output of the The CANFIS model combines fuzzy input with a
neural network given by: modular neural network to rapidly explain poorly
defined intricate functions using a basic component
ܱ ସ ǡ௝ ൌݓതതതതߤ ൌݓതതതതሺܲ ܼ ൅ܲ ܼ ൅ڮ ܲ ܼ ଶ௡ ൅ of fuzzy axon which applies a membership function
ఫ
௃௡
ఫ ௝
௃ଶ
ଶ
௃ଵ ଵ
ݍ ሻ ݂݋ݎ ሺͳ൑݆൑݉ሻ ሺͳͲሻ (MF) to the input [15].
௝
where ݓതതത is the output of layer 3, ሺܲ ൅ܲ ൅ 2.2 Configuration of the CANFIS network
௃ଵ
௃ଶ
ఫ
ڮ ܲ ሻ is the parameter set or consequent In modelling the five-input five-output CANFIS
௃௡

© International Telecommunication Union, 2018 75