The purpose of this study is to evaluate application of a type of supervised machine learning model called support vector machine (SVM) to freeway automatic incident detection. Many automatic incident detection algorithms are focused on identifying changes in traffic patterns but do not adequately investigate similarities in patterns observed under incident-free conditions. The most challenging part of real-time incident detection is recognition of traffic pattern changes when incidents happen during rush hour stop-and-go conditions. Incident detection can be described as a pattern classification problem and SVMs have pattern learning algorithms that have been successfully applied to incident detection. Previous evaluation studies have been based on either simulation data or the I-880 database. The possible issue with these is that non-incident traffic patterns may be biased by actual incident data. This study uses field traffic pattern data to overcome the problem of incident detection during peak hour. Data collected by the Dallas traffic control center including upstream and downstream speed and volume and typical upstream speed profiles. All parameters were used as base model input and different scenarios were defined, in terms of SVM kernel functions (the sigmoid and RBF) and different parameters combination. Cross-validation has been applied to increase classification accuracy. Based on this evaluation, the proposed SVM model provides reliable results.