Title | Chronic Kidney Disease Prediction Using Two Layer Adaptive Neuro-Fuzzy Inference System Topology |
---|---|
Author | J. Ijcsis |
Pages | 10 |
File Size | 883.3 KB |
File Type | |
Total Downloads | 94 |
Total Views | 742 |
International Journal of Computer Science and Information Security (IJCSIS), Vol. 14, No. 8, August 2016 Chronic Kidney Disease Prediction Using Two Layer Adaptive Neuro-Fuzzy Inference System Topology Mohammad Saber Iraji* Faculty Member of Department of Computer Engineering and Information Technol...
Accelerat ing t he world's research.
Chronic Kidney Disease Prediction Using Two Layer Adaptive NeuroFuzzy Inference System Topology Journal of Computer Science IJCSIS
Related papers
Download a PDF Pack of t he best relat ed papers
Fake Banknot es Classificat ion Using Wavelet and Anfis Topology Based on Vision Journal of Comput er Science IJCSIS GA_ RBF NN: a classificat ion syst em for diabet es Dilip Kumar Choubey Comparison of adapt ive neuro-fuzzy inference syst em and art ificial neut ral net works model t o cat eg… Ismail Saiboon
International Journal of Computer Science and Information Security (IJCSIS), Vol. 14, No. 8, August 2016
Chronic Kidney Disease Prediction Using Two Layer Adaptive Neuro-Fuzzy Inference System Topology Mohammad Saber Iraji* Faculty Member of Department of Computer Engineering and Information Technology, Payame Noor University, I.R. of Iran
Abstract— nowadays, many people suffer from kidney disease or die suddenly as a result of it. Chronic Kidney Disease Prediction is an important factor in the safety area. Neural networks, regression and fuzzy logic are useful tools for classification and prediction. Chronic Kidney Disease was estimated based on 24 features in 2015 and best results were obtained using neural networks approaches with 99.33 accuracy on 450 Sample. We presented in this paper proposed approach to predict Chronic Kidney Disease class than the actual class and calculate correct classification index, through two-layers adaptive neuro fuzzy system (TLA-ANFIS) based on 24 features and accuracy was achieved 100%. Finally, our results proves the impact of proposed TLA-ANFIS approach for Chronic Kidney Disease prediction in comparison to neural networks and regression methods.
Support Vector Machine (SVM) , and Regression from experimental data on the Chronic Kidney Disease[2,3,4]. L.JerlinRubini and P.Eswaran (2015) suggested a chronic kidney disease data set including 400 samples. They applied three classifiers multilayer perceptron, radial basis function network and logistic regression for classification into two categories CKD and not- CKD. Their Results reported that MLP was more accurate in predicting chronic kidney disease. In their research, there were no rules for the CKD category [1]. Ruey Kei Chiu and et al. (2013) used artificial neural networks including back propagation network (BPN), generalized feed forward neural networks (GRNN), and modular neural network (MNN) combined with genetic algorithm for the chronic kidney disease detection based on Biochemistry features and effective factors derived from Glomerular Filtration Rate (GFR) computation. Their investigations revealed that pure bp method predicted accurately on 430 patients. They proposed their model to test self-detecting publicly [5]. Abdurrahim Akgundogdu and et al. (2009) have presented Adaptive Neuro-Fuzzy Inference System (ANFIS) to predict the Renal Failure Disease. They modelled, their proposed system with seven features, three rules and implemented for 112 patients. The authors concluded that classification error index for ANFIS was less than Support Vector Machine (SVM) and Artificial Neural Networks (ANN) .Also they Stated that Anfis is a promising method for the diagnosis of kidney disease [6]. Vijayarani and Dhayanand(2015) Developed SVM and ANN to predict the kidney disease. They used data set containing 584 instances and six attributes including Age, Gender, Urea, Creatinine and GFR from kidney function test (KFT) dataset. Their results had reported that SVM achieved good results in terms of time and ANN in terms of classifying [7]. This paper is extracted from a project funded by the Payame Noor University has been adopted as “Chronic Kidney Disease Prediction Using Two Layer Adaptive Neuro-Fuzzy Inference System Topology”. Past researches prove that ANFIS has good performance in predicting. Our goal is designing an overall framework and it consists of all effective features for Chronic Kidney Disease Prediction using ANFIS. The large number of Adaptive Neuro fuzzy Inference
Keywords—chronic kidney disease, fuzzy neural network, neural network, multi -layer ANFIS
I. INTRODUCTION Kidneys are a double of limbs settled to lower back on either side of the body backbone. Kidneys remove toxins from the body through urine. Kidney failures are difficulty passing urine, loss of blood flow to the kidneys which cause kidney damage. Four of common kidney disease are Chronic Kidney disease, Acute Nephritic Syndrome and Chronic Glomerulonephritis and Acute Renal Failure.When the kidneys lose their function of filtering blood, the body will be stuffed of toxins and led to kidney failure and ultimately Death [1]. Chronic kidney disease (CKD) is associated with the following symptoms. • High blood pressure • Anemia • Coronary artery Disease • Diabetes mellitus • Dynasty history of kidney disease. • Leakage of Sodium and Potassium in blood Bacteria and albumin in urine So estimating the disease plays an important role in the survival of people. Blood and urine tests and ultrasound can be used to assess kidney function. It can be predicted by performing artificial intelligence techniques such as Artificial Neural Network (ANN),
220
https://sites.google.com/site/ijcsis/ ISSN 1947-5500
International Journal of Computer Science and Information Security (IJCSIS), Vol. 14, No. 8, August 2016
neurons is done by middle (hidden) layer. Neural network output is measured from the equation 1.
System (ANFIS) input features are major challenges using ANFIS and it is not applicable, with increased parameters. We presented a solution for many input features with solving modular problems for and we created a multi-layer architecture of SUB-ANFIS (MLA-ANFIS) for Chronic Kidney Disease Prediction with 100% accuracy.
=
=
+ exp − − We utilised feed forward neural network and BP( Back propagation) learning method in this research(Figure1). More information about neural networks is available in [11].
II.METHODOLOGY AND MODEL Neural networks, fuzzy systems, genetic algorithms are soft computing techniques that are frequently used in systems modelling and solving them. Fuzzy logic stated by Lotfi Zadeh[8], is a versus binary logic and provides new ways to solve problems. Neuro-fuzzy systems are fuzzy systems that are used in ANN in order to determine fuzzy sets and fuzzy rules from training data. Adaptive neuro-fuzzy inference system (ANFIS) is a special type of fuzzy neural network, which is used in modelling nonlinear systems. Training data, is a model and demonstration for different systems and ANFIS discovers the exact parameters of the membership function according to experimental data. The ANFIS learns Patterns from the data set and tunes the system parameters according to a given error criterion [9]. Successful applications has been reported in the ANFIS system for biomedical engineering and classification [7,10]. Neural networks are made of neurons and they are used for predicting the relationship between inputs and output. The relationship between input and output inputmf
=output of neuron j of hidden neuron =input i to hidden neuron =weight connection among input and hidden neuron from input i to neuron j =bias of hidden neuron j = transfer function for hidden neuron j Transfer function is following:
First We Mentioned about Chronic Kidney Disease Prediction and we need it, in this paper. Then we pay attention to many papers in this field. Data have been considered as the base of the research from paper [1] of 400 sample due to a mistake values, 158 correct sample was selected and we are trying to estimate the Chronic Kidney Disease classification more accurately by neural network and anfis methods.
input
+
Figure 1. Neural network structure
Neural networks deploy a model of system by using such a data that 70% of it is randomly selected for training data, and 15% for validation and testing. ANFIS (Adaptive Neuro Fuzzy Inference System) is based on sugeno [12,13]. A generic rule in a Sugeno fuzzy pattern has the form of If Input 1 = p and Input 2 = q, then output is z = ap + bq + c. Figure 2 depicted the anfis neural network [14]. rule
outputmf
output
Logical Operations and or not
Figure 2. Adaptive Neuro fuzzy Network (anfis)
221
https://sites.google.com/site/ijcsis/ ISSN 1947-5500
International Journal of Computer Science and Information Security (IJCSIS), Vol. 14, No. 8, August 2016
In Figure 2 first layer is the degree of membership of linguistic variables. The second layer is 'rules layer'. After the linear composition of rules at third layer, then we assign the degree of belonging to a special class by Sigmund's function in layer 4. ANFIS is a type of fuzzy neural network with a learning algorithm based on a set of training data for tuning an available rule base from the training data. Data used CDK data set from UCI and given training data to ANFIS the related rule is set, and obtain more accurate output (Figure 2).These features are defined below.
age specific gravity hypertension pedal albumin
1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 16. 17. 18. 19. 20. 21. 22. 23. 24.
Anfis 1
age - age bp - blood pressure sg - specific gravity al - albumin su - sugar rbc - red blood cells pc - pus cell pcc - pus cell clumps ba - bacteria bgr - blood glucose random bu - blood urea sc - serum creatinine sod - sodium pot - potassium hemo - hemoglobin pcv - packed cell volume wc - white blood cell count rc - red blood cell count htn - hypertension dm - diabetes mellitus cad - coronary artery disease appet - appetite pe - pedal edema ane - anemia
Anfis 2
Anfis 9
sugar appetite blood pressure
Anfis 3
bacteria blood glucose serum creatinine
Anfis 4 Final Anfis
sodium potassium hemoglobi
Anfis 5
Anfis 10
packed cell volume pus cell
One limitation of anfis is when the number of input variables are high, with face memory shortage [15]. To this end, we have used the following topology (Figure 3).
red blood cells red blood cell count
III.EXPERIMENTAL RESULTS We implemented our proposed system in MATLAB version 7.12 on Laptop, 1.7 GHZ CPU, and we used the roots mean square error (RMSE) indicators, In order to determine the number of hidden layer neurons. = ∑
=
∑
−
−
Anfis 6
hypertensi diabetes mellitus blood urea
(3)
Anfis 7
Anfis 11 coronary artery disease
(4)
anemia
Anfis 8
white blood cell
Figure 3. Two layer anfis topology model for Chronic Kidney Disease Prediction
222
https://sites.google.com/site/ijcsis/ ISSN 1947-5500
International Journal of Computer Science and Information Security (IJCSIS), Vol. 14, No. 8, August 2016
is evident that MSEs For the best anfis system performance with training and validation data are 0.0000, 0.0028.
result rmse for neural network 0.5 train test validaion
0.45 0.4
mse error curve 3
0.35
RMSe
0.3
train nn validation nn test nn train anfis validation anfis
2.5
0.25
2 MSe errors
0.2 0.15 0.1 0.05 0
1 0
5
← train
10 15 20 25 30 35 40 number of hiden neurons and best hiden neuron=38
45
50
0.5
Figure 4. (a) RMSE between actual class and predicted class using neural network for determine number of hidden neuron
0
Figure 4 shows RMSE between actual class and predicted class using neural network from 1 to 50 hidden neuron. As it's seen in Figure, the optimum number of hidden neurons are 38 with least RMSE. Neural network architecture 24-38-1 was considered and reached its best performance after five epochs.Figure 5 shows the amount of MSE for training validation, testing data using best topology of neural network. For the optimum network performance at epochs, five Mse for training and validation data is 0.0000 and 0.0245 in figure 5. test nn validation nn train nn
MSe errors
1.5
1
Correlation coefficient demonstrated predicted class and for training and validation data actual class namely using anfis in figure 8.
0.5
1
2
←
3 4 5 Epoch numbers=6best performance=5
6
7
The is compared in table1 and neural network, regression and proposed anfis system are used. The for training data which uses anfis system is 1 that proves the anfis system has better performance than neural network in validation the training phase. Finally in comparison to data, the anfis ultimately performs better with =0.9943 versus =0.9251, 0.98872 using neural network and regression it is suggested. Figure 9 shows deviations of predicted Chronic Kidney Disease from actual (Dev%) for sample records. Information’s Value for Chronic Kidney Disease records has been presented in table 2 (uci data set).
Best Validation Performance is 0.024508 at epoch 5 0
10
-5
10
-10
10
test nn validation nn train nn best
-15
10
0
1
2
3
4
5
7
In proposed Anfis system a database of 158 records was considered, In order to train and test the fuzzy neural network. After calculating 24 features were described above for 158 records, 110 records were considered for training Anfis and 24 record Was Allocated to Test and Validate system. After setting network parameters to generate fis =grid partition, optim.mrthod=hybrid, linier, training fis epochs=6, gaussmf membership function with two mf Rmse(Root mean squared errors ) for training data Obtained 0.0000.
2
0
2 3 4 5 6 Epoch numbers6best performance nn= 5,anfis=1
Figure 7 shows the correlation coefficient, according to for the the predicted class and actual class Wordy training, and validation data using Regression and neural network by Chronic Kidney Disease Prediction.
Training (red) and validation (green),test(blue) error curve
2.5
1
Figure 6. Comparison training, validation and test error with neural network and anfis
3
Mean Squared Error (mse)
1.5
6
6 Epochs
Figure 5.(a)Training ,validation and test error curve (b)best performance for neural network with validation data is epochs 5
Figure 6 shows the deal of MSE for the training validation, testing data using neural network and anfis. It
223
https://sites.google.com/site/ijcsis/ ISSN 1947-5500
International Journal of Computer Science and Information Security (IJCSIS), Vol. 14, No. 8, August 2016
Data Fit Y = T
Output ~= 0.96*Target + 0.012
1
Neural network
0.8
Train
valiation
1.0000
0.9251
Anfis 3-8
1.0000
0.9943
Regression
0.97849
0.98872
0.6
0.4
Rsqure for train data with anfis topology =1 1
0.2
Data Fit Y= T
0.9
0
0.2
0.4
0.6
0.8
Output ~= 1*Target + 7.9e-010
0 1
Target
Rsqure for validation data with regression=0.98872 1.2 Data Fit Y = T
1
Output ~= 1.1*Target + -0.001
using neural network, regression and ANFIS
Table 1. Comparison
Rsqure for train data with regression=0.97849
0.8
0.8 0.7 0.6 0.5 0.4 0.3 0.2
0.6
0.1 0.4
0
0.2
Target Rsqure for validation data with anfis topology =0.99431
0
0.2
0.4
0.6
0.8
1
1
0.2
0.4
0.6
0.8
1
Output ~= 0.92*Target + 1.6e-006
0
1.2
Target Rsqure for train data with neural network =1 1 Data Fit Y = T
Output ~=1*Target +9.7e-006
0.9 0.8 0.7 0.6 0.5
0.8 0.7 0.6 0.5 0.4 0.3 0.2
0.4
0.1 0.3
0
0.2
0
0
0.2
0.4
0.6
0.8
1
Target
0.1
0
0.2
0.4
0.6
0.8
Figure 8. Plot predicted class and actual class R for (a) train (b) validation data using anfis system
1
Target Rsqure for validation data with neural network =0.92515
dev% for total data with ...
100
Data Fit Y = T
1.2
neural network anfis regression
50
1
0
0.8
-50 Dev%
Output ~=0.89*Target +0.0003
Data Fit Y=T
0.9
0
0.6
-100 -150
0.4
-200 0.2
-250 0 0
0.2
0.4
0.6
0.8
1
-300
1.2
Target
Figure 7. Plot predicted class and actual class R for (a) train (b) validation data using Regression, (c) train (d) validation data using neural network
0
20
40
60
80 100 Record number
120
140
160
Figure 9. Comparison predicted Chronic Kidney Disease by neural network, Regression and anfis
224
https://sites.google.com/site/ijcsis/ ISSN 1947-5500
International Journal of Computer Science and Information Security (IJCSIS), Vol. 14, No. 8, August 2016
Table 2. Influence factors for Chronic Kidney Disease classification features
min
Correct classification% Incorrect classification% RMSE
mean
age - age
6
83
49.5633
bp - blood pressure
50
110
74.0506
sg - specific gravity
1.005
1.025
1.0199
al - albumin
0
4
0.7975
su - sugar
0
5
0.2532
rbc - red blood cells
0
1
0.8861
pc - pus cell
0
1
0.8165
pcc - pus cell clumps
0
1
0.0886
ba - bacteria
0
1
0.0759
bgr - blood glucose random
70
490
131.3418
bu - blood urea
10
309
52.5759
sc - serum creatinine
0.4
15.2
2.1886
sod – sodium
111
150
138.8481
pot – potassium
...