Adaptive Learning Algorithms for Personalized Skill Development

Abstract

Personalised skill development in professional and vocational contexts demands adaptive learning algorithms that model individual learner trajectories across heterogeneous skill graphs, optimise content sequencing for transfer and retention, and update dynamically as learner performance evolves. Unlike structured academic curricula, professional skill development involves open skill graphs with complex prerequisite structures, diverse learner prior knowledge, and performance objectives tied to real-world task completion rather than test scores. This paper proposes the Adaptive Skill Development Algorithm Framework (ASDAF), evaluating five adaptive learning algorithms -- Bayesian knowledge tracing with skill graphs (BKT-SG), deep reinforcement learning curriculum (DRL-C), collaborative filtering recommendation (CFR), item response theory adaptive testing (IRT-AT), and graph neural network skill modelling (GNN-SM) -- across 24 skill development benchmarks in software engineering, data science, and project management professional domains. ASDAF evaluates 3,600 learner trajectories from corporate learning platforms over 6 months. Key results: DRL-C achieves the highest Skill Development Efficiency Index (SDEI = 0.902), reducing time-to-proficiency by 32.4% versus linear curriculum; GNN-SM achieves 94.2% skill prerequisite prediction accuracy; BKT-SG achieves 88.6% mastery assessment accuracy. The framework provides algorithm selection guidance and open benchmark tools for adaptive professional learning design.