Subcellular localization problem has been researched to discover genes’ function and discover putative circulating biomarker proteins. Secreted proteins include extracellular matrix proteins which surround cells and influence critical cell behaviors. Extracellular matrix proteins are related to cancer progression and can be a therapeutic target. However, there was no paper to predict extracellular matrix proteins computationally. Introducing new features considering characteristic sequence repeats of extracellular matrix proteins improved the accuracy of prediction. Also 17 distinctive features are found which are robust to classifiers by analyzing feature important values given by random forest. Overall 98 % accuracy was achieved from 5-fold cross validation. Finally 30 novel extracellular matrix proteins are predicted using 17 features.