A comprehensive dataset of Infant Facial Expressions of Pain Intensity

Facial expressions are critical for interpreting and diagnosing internal emotional states, particularly in clinical contexts. Recognition and accurate assessment of pain-related facial expressions are essential for effective patient care. Existing datasets primarily represent adult pain expressions, with comparatively fewer resources dedicated to infant pain. This study introduces the Infant Facial Expressions of Pain Intensity (IFEPI) dataset, which categorizes infant pain into four levels: no pain, mild pain, moderate pain, and severe pain. The dataset comprises 25,000 spontaneous images collected from 120 infants at two hospitals in Khyber Pakhtunkhwa, Pakistan, both affiliated with the Health Department of the Government of Pakistan and the World Health Organization (WHO). The methodology encompasses environment setup, image acquisition, preprocessing, and image refinement. The IFEPI dataset is currently the largest available dataset for pain-related facial expressions. To validate the dataset, four machine learning models–logistic regression, support vector machine (SVM), k-nearest neighbors (KNN), and random forest–were employed, alongside eight deep learning models: LeNet, AlexNet, Visual Geometry Group-16 (VGG-16), Inception Version 3 (InceptionV3), Densely Connected Convolutional Network-121 (DenseNet121), Densely Connected Convolutional Network-201 (DenseNet201), Efficient Convolutional Neural Network-B0 (EfficientNetB0), and Efficient Convolutional Neural Network-B7 (EfficientNetB7). DenseNet201 achieved training and testing accuracies of 93.37% and 94.95%, respectively, while EfficientNetB7 achieved training and testing accuracies of 96.82% and 94.61%, respectively.