Artificial neural network based particle swarm optimization in predictions mortality rate of broiler chicken

Artificial neural network based particle swarm optimization in predictions mortality rate of broiler chicken

1. why they should make this journal?

Success in broiler farms can be assessed from the mortality rate of chickens and also

a solution to reduce the mortality rate of chickens. But in the process there are obstacles that is

the death of hickens that tend to fluctuate so that of course resulted in financial losses for farmers.

To prevent it required a method that can predict and control the mortality rate of broiler chickens.

In this research, data mining method used is Artificial Neural Network (ANN) based on Particle

Swarm Optimization (PSO), to produce accurate prediction with excellent iteration and also

small error rate. The results of analysis in predicting mortality rate of broiler chickens by

artificial neural network method in combination with Particle swarm optimization get better

RMSE result (0.135) than Artificial Neural Network have not been optimized (0.381) with

january mortality data and result of application quisioner made value 83,125 which can be

categorized well and enough help the farmer in controlling prediction of chicken mortality rate

broiler.

2. Methodology

Methodology

In this research used two methods that support each other is Particle Swarm Optimization Neural

Network.

3.1. Step one with Artifical Neural Network [9]

For each input unit (xi, i=1,2,3,…n) receive signal xi and pass the signal on the layer above it (hidden

layer). Every hidden layer (zj, j=1,2,3,…,p) adds weighted input signals:

_ = 0 + ∑ (1)

Use the activation function to calculate the output signal:zj = f (z_inj)and send the signal to all units in

the top layer (output units). For each output units (yk, k=1,2,3,…,m) calculate weighted input signals”

_ = 0 + ∑ (2)

Use the activation function to calculate the output signal.yk= f (y_ink)and send the signal to all units in

the top layer (output units). For each output unit (yk, k=1,2,3,…,m) receiving target patterns connected

with the learning input pattern. Then calculate the error information.

σk = (tk – yk) f’(y_ink) (3)

then calculate the weight correction (which will be used to fix wjkvalue) :

Δwjk = α σk zij. (4)

also calculate the bias correction (which will be used to fix w0kvalue) :

Δwjk = α σk. (5)

send σk to the units in the bottom layer. Each hidden units (zj, j=1,2,3,…,p) calculate the delta units

(from units in the top layer):

_ = ∑ (6)

multiply this value withderivativefromits activation function to calculate information of error:

σj = σ_inj f’(z_inj) (7)

also calculate the bias correction (which will be used to fix v0jvalue):

Δvjk = α σj xi (8)

also calculate the bias correction (which will be used to fix v0jvalue):

Δv0j = α σj. (9)

Each output unit (yk, k=1,2,3,…,m) improve bias and weight (j=0,1,2,3,…,p). wjk (new) = wjk(old) +

Δwjk for each hidden layer (zj, j=1,2,3,…,p) improve bias and weight (i=0,1,2,3,…,n)

Test condition stopped [10].

3.2. Step two with particle swarm optimization

Where: X = particle position, V= particle velocity, w= weight of inertia, c1, c2 = acceleration coefficient,

P= number of particles in swarm. The values of r1 and r2 are arranged so that the random value is

intended to provide stochastic properties in the cognitive component and social components This

stochastic nature causes each particle to move in a semi random way, strongly influenced in the direction

of the best solution of the particles and the best globalsolution of swarm [11].

3.3. Step three software testing with SQA

There are three points of understanding of software quality, including as follows: (1) software

requirements are the foundation from whichquality is measured; (2) specific standards that define

development criteria thatguide the manufacture of a software; (3) there are implicit needs often

overlooked (eg, desire for the best maintenance); and (4) scores of responder = <audibility score> * 0.10

+ <scoreaccuracy> * 0.10 + completeness score> * 0.15 + <errortolerance score> * 0.10 + <score

execution efficiency> *0.10 + <operability score> * 0.15 + < Simplicity score> *0.15 + <learning score>

* 0.15 Evaluation based on criteria /average score of respondents [12].

3. resolution from this journal 

For accuracy and error resulting from the Neural Network method in predicting the mortality rate of

chicken boiler, it was 5,854 (december). While the average RMSE results obtained from the method of

artificial neural networks which is 6,222 each month. Existing data cannot affect the RMSE results

obtained as evidenced by the results of the RMSE obtained in September is the highest result that is

equal to 14.115 while the lowest results can be seen in June that is equal to 0.730. While the RMSE

results obtained from the method of artificial neural networks with backpropagation models are

enhanced by particle swarm optimization which is 2,032 (december). For the average RMSE obtained

is equal to 1.889 better by 4.3% of the neural network method with the backpropagation model before

being optimized. With the highest RMSE of 7.119 and the lowest RMSE of 0.223. And by using an

application that can run the Neural Network method based on Particle Swarm Optimization and with the

results of predictions with a good level of error and accuracy and the results of the application test from

the management to the audience with the SQA results obtained that is equal to 83.125 this shows that

the management is helped with this application and also take the steps needed to prevent mortality of

broiler chickens from the prediction of ANN and PSO methods.