Enhancing Facial Expression and Micro-Expression Recognition Through Hybrid Dimensionality Reduction and Machine Learning Classification

Abstract

Facial expression and micro-expression recognition have become essential research topics in affective computing, computer vision, and intelligent human–computer interaction. Despite recent advances, accurately recognizing subtle facial movements remains challenging because of their short duration, low intensity, and the high dimensionality of facial image data. This study proposes a hybrid machine learning framework that combines advanced dimensionality reduction techniques with supervised classification algorithms to improve recognition performance while reducing computational complexity. The proposed approach employs geometric facial landmark extraction using the Dlib library, followed by image preprocessing through grayscale conversion, histogram equalization, and normalization. Three dimensionality reduction techniques—Kernel Principal Component Analysis (KPCA), t-distributed Stochastic Neighbor Embedding (t-SNE), and Uniform Manifold Approximation and Projection (UMAP)—are investigated to generate compact and discriminative feature representations. These reduced feature spaces are subsequently classified using Support Vector Machine (SVM), Decision Tree, and Random Forest algorithms. Experimental evaluation was conducted using the AffectNet and CASME II datasets, which provide diverse facial expressions and spontaneous micro-expressions. The findings demonstrate that dimensionality reduction substantially decreases computational cost while preserving discriminative information. Among the evaluated combinations, Random Forest integrated with KPCA and UMAP achieved the highest classification accuracy of approximately 94%, while simultaneously reducing training time by nearly 40%. The results indicate that combining nonlinear dimensionality reduction with ensemble learning provides an effective and computationally efficient solution for facial emotion recognition. The proposed framework offers promising applications in intelligent surveillance, healthcare, psychological assessment, and next-generation human–computer interaction systems.