Wireless Capsule Endoscopy (WCE) is an important device to detect abnormalities in small intestine. Despite emerging technologies, reviewing capsule endoscopic video is a labor intensive task and very time consuming. Computational tools which automatically detect informative frames and tag abnormal conditions such as bleeding, ulcer or tumor will dramatically reduce the clinician's effort. In this paper, we explored various machine-learning methodologies based on different feature extraction and selection criteria, and developed an optimized classification method. The experiment results shows that, comparing to texture feature, using color feature for classification achieved better accuracy, regardless of machine-learning method chosen. The proposed method has been applied in real data taken from capsule endoscopic exams. For informative frames detection, classification method using color feature gives an accuracy of 94.10% and 93.44% for support vector machines (SVM) and neural network (NN) classifiers respectively. For the bleeding detection using color feature, the accuracy achieved 99.41% and 98.97% for SVM and NN respectively. In addition, we also investigated the computational time required for feature extraction and classification. In our experiments, color feature significantly outperformed texture feature in WCE image classification. The overall computational time (per frame) using color feature is 0.7125s (informative frame with SVM), 1.0329s (informative frame with NN), 0.51s (bleeding frame with SVM) and 1.2163s (bleeding frame with NN). Classifiers for more gastro-intestinal (GI) diseases detection will be developed based on this work subsequently.