Prediction of DNA Hepatitis C Virus based on Recurrent Neural Network-Back Propagation through Time (RNN-BPTT)

Hepatitis C Virus (HCV) is one virus that has a high mutation rate in the world. To predict the mutation can be used fundamental analysis and technical analysis. Fundamental analysis relies on external factors such as attributes attached to the primer and the isolate. While technical analysis learns the movement of the mutation itself by relying on graphs and mathematical formulas. This study combines fundamental analysis and technical analysis in predicting HCV mutations. Application of Recurrent Neural Network (RNN) method as a form of technical analysis and fundamental analysis is applied in the form of including some fundamental factor data as training datasets. RNN is a neural network that has a feedback connection to the neuron itself, or a previous neuron. RNN is able to reactivate actual data values in the past to be re-entered with actual data values at the moment. This study used Elman network architecture with Back Propagation Through Time learning algorithm (BPTT) and used Linear normalization. The problem when predicting HCV mutations is how to determine the best learning rate value. Therefore the fundamental approach will also be incorporated into the neural network. On the backward process, to calculate the value of weight correction is to multiply the value of learning rate with hidden neuron value. We propose that each neuron has an adaptive learning rate according to the condition of the neuron. Where each input neuron from this study is the result of the HCV primer normalized in preliminary research. Each primer has attributes that can be developed as decision support. Test results show the smallest error of the prediction process by RMSE are 0.0163000, with accuracy prediction value are 99%.