Data Analysis and Machine Learning on Eye-Tracking Data to Interpret Consumer Behavior for Yogurt Products with a Novel Edible Bio-Film

This study presents a data-driven framework that integrates eye-tracking technology with algorithmic analysis to uncover patterns in consumer attention and preference. Using yogurt packaging as a case study, we recorded the gaze behavior of 100 participants as they viewed paired images, one featuring conventional plastic film and the other a novel, brown edible film derived from microorganisms. A key methodological innovation lies in the unsupervised identification of Areas of Interest (AOIs): Instead of relying on pre-defined regions, we employed clustering algorithms (e.g., k-means with silhouette score optimization) to detect natural groupings in gaze data, enabling more objective and personalized AOI boundaries. Multiple eye-tracking metrics were analyzed, including fixation duration, saccade duration and velocity, and pupil diameter. These features were processed and modeled using a suite of machine learning classifiers, including Decision Tree, Random Forest, Gaussian Process, Multilayer Perceptron, and AdaBoost. The results demonstrated classification accuracies of up to 83% in predicting participants’ product preference based solely on visual behavior, highlighting a strong correlation between gaze dynamics and consumer decisions. Statistical comparisons also revealed significantly longer gaze durations toward the edible film, suggesting increased attention or a novelty effect. Future work should validate these findings by combining eye-tracking results with sensory evaluation methods, such as external preference mapping, and physicochemical analyses of the products to better understand how visual attention relates to sensory perception and product characteristics. The proposed approach showcases how advanced computational methods can enhance the interpretation of eye-tracking data in consumer research. Given the focus on yogurt, this methodology holds significant potential for informing sustainable packaging development, consumer segmentation, and product positioning in dairy and related food sectors. While the stimulus involved sustainable packaging, the broader contribution lies in demonstrating how machine learning and algorithmic analysis can uncover latent drivers of consumer choice from complex biometric signals.