Research Paper
Year: 2020 | Month: February | Volume: 7 | Issue: 2 | Pages: 120-128
Performance Analysis of Adaptive Neuro-Fuzzy Inference System (ANFIS) With Subtractive Clustering In the Classification Process
Ahmad Rizal1, Tulus2, Syahril Efendi2
1Postgraduate Students at Universitas Sumatera Utara, Medan, Indonesia
2Postgraduate Lecturer at Universitas Sumatera Utara, Medan, Indonesia
Corresponding Author: Ahmad Rizal
ABSTRACT
Adaptive neuro-fuzzy inference system (ANFIS) is a combination of artificial neural network and fuzzy inference system (FIS) techniques. ANFIS is used to significantly improve the results of the classification process, while substractive clustering is a method of grouping based on the potential density of the data to calculate the cluster center to the surrounding data points. With substractive clustering, large datasets are divided into groups and determined radii are also called cluster radii. The cluster radius serves to determine the area of influence that has a value between 0 and 1. The results of this study use the FIS substractive clustering parameter to minimize mean square error (MSE), root mean square error (RMSE), mean absolute error (MAE) with maximum epoch 1000, target error=0.01. Testing training data 60% of the dataset values, MSE=192750.4593, RMSE=439.0336 with epoch 369, and MAE=0.00012481 with epoch 94, radius=0.3 squash factor=1.5, accept ratio=0.7, reject ratio=0.35. Testing data testing 40% of the MSE dataset=205013,853, RMSE=452.7846 at epoch 49 and MAE=2.33968 at epoch 51, radius=0.9 squash factor=1.25, accept ratio=0.5, reject ratio=0.25. Testing uses 70% training data and 30% testing data from the dataset. In the 70% training data test, the minimum MAE value=0.00010119 with the radius value=0.2, accept ratio=0.3, reject ratio=0.15 and squash factor=1. 30% testing data test obtained the MAE drinking value=0.060341 with a radius value=0.1, accept ratio=0.7, reject ratio=0.35 and squash factor=1.5.
Keywords: Adaptive Neuro-Fuzzy Inference System, Substractive Clustering, Mean .Square Error, Root Mean Square Error, Mean Absolute Error
[PDF Full Text]