<![CDATA[This study proposes a multi-class classification framework for monitoring blockchain system performance as a step toward integration within blockchain-based learning management systems (LMS). Reliable performance monitoring is essential because smart contracts in educational settings depend on timely and accurate system responses to ensure valid grading and credential issuance. A dataset of 3,081 transactional logs was generated from simulated blockchain testbed, capturing throughput, latency, block size, and send rate. Throughput values were discretized into seven qualitative categories ranging from “Very Poor” to “Very Good” using quantile-based binning. Preprocessing involved data cleaning, categorical encoding, Z-score normalization, and label encoding to ensure model compatibility. Five algorithms: Logistic Regression, Decision Tree, Random Forest, Support Vector Machine (SVM), and K-Nearest Neighbors (KNN) were trained and evaluated using stratified 80–20 partitioning and 5-fold cross-validation with grid search for hyperparameter tuning. Performance metrics included accuracy, macro precision, recall, and F1-score. Random Forest achieved the best results with 91.35% accuracy, 0.910 macro precision, 0.911 recall, and 0.910 F1-score, outperforming other models by handling complex feature interactions and reducing variance. Decision Tree offered strong interpretability (88.32% accuracy), while Logistic Regression (84.97%) and SVM (84.86%) provided stable performance. KNN showed balanced results (87.78%) but incurred high computational costs. The findings demonstrate that multi-class stratification provides more actionable insights than binary methods, supporting low-latency decision-making for smart contract execution in decentralized LMS ecosystems. The novelty of this research lies in applying multi-class classification instead of binary methods, enabling nuanced monitoring. Future work will validate the framework in real blockchain-LMS deployments.]]>
