Analisa Performa Metode Training Artificial Neural Network dalam Mendeteksi Total Harmonic Distortion

Masyitah Aulia
Ony Ramadhan Armanto (Universitas Teknologi Surabaya)
Moh.Yasya Bahrul Ulum (Universitas Teknologi Surabaya)

Hexagon
Vol 6 No 2 (2025): Vol. 6 No. 2 (2025): HEXAGON - Edisi 12

This Study aims to asses the performance of several training algorithms in Artificial Neural Network for pattern classification task based on input data. The research problem focuses on how training algorithms can produce low error rates in classification. The scientific gap lies in the limited comparative studies that comprehenseively exmine various training methods with variations in the number of hidden neurons. This research utilizes five training methods, Levenburgh – Marquadt, Scaled Conjugate Gradient, Resilient Backpropagation, BFGS Quasi – Newton, and Bayesian Reqularization. The hidden neuron variations range from 5 to 50 neurons. The result indicate that the best performance is achieved using the Bayessian Regularization method, which produces a MAPE value 0.01, an RMSE of 0.0167, abnd accuracy of 99.98%. These findings demonstrate that choosing the appropriate training function significantly affects the performance of Artificial Neural Networks.