Combining YOLO and Scikit-Learn Improves Real-World Audience Classification

This study addresses the problem of automatic object classification by leveraging the strengths of both deep learning and traditional machine learning. The main goal of this project is to develop a prototype application capable of efficiently and accurately recognizing and classifying objects in images. To tackle this, the YOLOv10 model for object detection was used, then extracted features such as bounding-box size [3] and average color. If an image is of poor quality or YOLOv10 fails to detect any object, this study applies PCA to enhance image quality. These extracted features are then used to train a Random Forest classifier from the scikit-learn library. The performance of the Random Forest classifier is optimized using GridSearchCV [2] and evaluated using StratifiedKFold [5]. The results showed that the YOLO + Random Forest combination system achieved an overall accuracy of 93%, with a higher average Precision and F1-score than using YOLOv10 alone. The combined model significantly improves the ability to classify glass and organic objects, although the number of samples of these two types is limited. The study concluded that the combination of YOLOv10 and Random Forest is an effective approach to building an automated object classification system, taking advantage of the detection speed of deep learning and the characterization-based classification capabilities of traditional machine learning, contributing to intelligent object management.