Hepatitis C treatment is currently facing many challenges, such as high costs of medicines, side effects in patients, and low success rates with Hepatitis C Virus genotype 1b (HCV-1b). In order to identify what characteristics of HCV-1b cause drug resistance, many sequence analysis methods are conducted, and bio-markers helping to predict failure rates are also proposed. However, the results may be imprecise when these methods work with a dataset having a small number of labeled sequences and short length sequences. In this paper, we aim to predict outcomes of the HCV-b treatment and characterize the properties of HCV-b by using the combination of a feature selection and semi supervised learning. Our proposed framework improves the prediction accuracy about 5% to 8% in comparison with previous methods. In addition, we obtain a set of good discriminative subsequences that could be considered as biological signals for predicting a response or resistance to HCV-1b therapy.