Due to the specific properties of time series, most standard machine learning classification methods in data mining do not work well for time series classification (TSC). So far, empirical studies have shown that the 1-nearest-neighbor method with Dynamic Time Warping distance (1-NN + DTW) is highly effective to classify directly time series data. Recently, there has been one research work by Kien and Anh that proposed a framework for TSC which uses motif information to create feature vectors for time series before using some standard machine learning classification methods. Experimental results in the work showed that all 1-nearest-neighbor, neural network (ANN) and support vector machine (SVM) with motif-based feature extraction can perform better than 1-NN with Euclidean distance in time series classification. In this paper, we enhance the above-mentioned framework by the main idea: after transforming time series into feature vectors, we apply Random Forest, an effective ensemble method for data classification on the feature vectors. Besides, we compare the performance of our proposed method with that of 1-NN + DTW, the golden baseline method in TSC. Results of extensive experiments on twenty-one benchmark datasets show that our proposed TSC method which is based on motif information outperforms not only 1-NN + DTW, the gold baseline algorithm, but also Naïve Bayes method, Decision Tree and SAX-VSM, a well-known feature-based method, in time series classification. These results indicate that our proposed method, an ensemble classifier significantly outperforms the individual classifiers.