Introduction: Evaluating multidimensional language competencies in tertiary-level Arabic education poses persistent methodological difficulties, as conventional scoring systems frequently reduce complex proficiency dimensions to single aggregated values that conceal underlying skill structures and diminish the instructional utility of assessment feedback. Research Objectives: The present study constructs an empirically grounded competency profiling framework by combining clustering algorithms with predictive modeling techniques to uncover latent proficiency patterns among Arabic language learners. Methodology: A cross-sectional quantitative design was adopted using data from 128 students in the Arabic Language Education program at Universitas Negeri Jakarta, whose scores across listening, speaking, reading, and writing skills were analyzed through Fuzzy C-Means (FCM) clustering to identify latent proficiency groupings, followed by the use of a feedforward neural network to model predictive relationships between individual skill domains and overall academic performance. Results: Three learner profiles emerged: low, moderate, and high proficiency each showing statistically significant inter-group differences across all skills (p < 0.001), with effect size estimates (η² = 0.20–0.25) confirming moderate to substantial cluster-level variance, while the neural network attained 93.33% accuracy with a minimal mean squared error (MSE = 2e⁻⁰⁶). Unique Contribution: This study offers an empirically validated hybrid framework synthesizing exploratory clustering with predictive analytics to advance language competency assessment methodology. Conclusion: Arabic language proficiency appears to be organized along a clearly delineated continuum that is statistically distinguishable and reliably predictable. Recommendations: Future research should incorporate broader learner-level variables and apply this framework across diverse educational settings to strengthen generalizability and instructional relevance.