Sentiment analysis for pain detection using hyperparameter optimization on SVM
Arhina Ghosh, Neha Tyagi · 2025
The current study introduces a new method of pain classification in videos based on combining both appearance and motion. Here, using deep learning and machine learning, a system that efficiently extracts and analyses visual features of frames in a video is designed in this work. Since working with an image dataset, the deep features are extracted by utilizing the pretrained VGG16 model, after which, PCA reduction is applied and chi-squared tests are conducted for feature selection. The selected features are then used as inputs to the support vector machine (SVM) classifier where the hyperparameter tuning is done using the GridSearchCV. Our system performs well in pain classification tasks because our system has a higher test accuracy than the other methods applied on MintPain dataset. This solution, which is a blend of both deep learning and machine learning, would provide a way forward for more accurate and reliable pain detection.