Recent technological advances in manufacturing allow for high-speed mass production and more manufacturers to utilize online inspection machines, such as cameras that detect defective products, to enhance their production ability. However, online inspection machines are still in the development stage and might cause misclassifications, which can result in unnecessary investment or potential loss in manufacturers' reliability. The objective of this study is to build a more precise defect detection system to discriminate between non-defective products and defective products. Previous studies utilized filtering methods such as feature extraction in relation to a filtering window before classification to enhance precision. Our study adopted a different approach in classification process to further increase its precision. We designed a preprocessing process and a feature extracting process based on a variable importance from Random Forest, and a classification process composed of our newly developed ensemble classifiers. As a result, we succeeded in constructing a defect detection system that performs classification with over 90% accuracy, which is more precise than the previous system. Making the most of AdaBoost's algorithm to overcome the difficulty in the classification resulted in an improvement in the system's precision.