Human footprint analysis has valuable applications across fields such as environmental monitoring, forensics, wildlife tracking, and medical diagnostics. Despite its importance, current models capable of accurately identifying and analyzing human footprints from raw images remain limited—especially those that offer both high precision and a user-friendly interface. This research presents a lightweight Android application, CrookFoot, which integrates a deep learning-based Convolutional Neural Network (CNN) to perform multi-output classification of barefoot images. The model predicts the user’s age group, foot type (flat, normal, or high arch), and calculates the arch index using a single image input. The CNN architecture comprises 12 layers and is optimized for mobile deployment using TensorFlow Lite. Trained on an augmented dataset of 100+ footprint images, the model leverages 17 distinct features—spanning geometric (5), GLCM texture (6), LBP texture (3), statistical (3), and classification-based metrics (2)—to deliver robust and detailed predictions. Comprehensive evaluation metrics such as confusion matrices, precision, recall, F1 scores, and ROC curve analysis validate the model’s performance. Designed for offline use, CrookFoot delivers real-time diagnostic results and personalized product suggestions, offering an affordable and accessible foot screening solution with potential for global healthcare integration.