Karunarathne, Lakmali ORCID: https://orcid.org/0009-0000-7720-7817 (2026) Machine Learning-based Query Performance Optimization in Large-Scale IoT Databases. Journal of Information Technology and Digital World, 8 (4). pp. 287-307.

Preview

Text Machine_Learning_based_Query_Performance_Optimization_in_Large_Scale_IoT_Databases.pdf - Published Version Available under License Creative Commons Attribution Non-commercial. Download (674kB) | Preview

Abstract

The rapid growth of Internet of Things (IoT) deployments generates continuous, high-volume telemetry that strains query processing in large-scale databases, and static rule-based or cost-based optimizers do not adapt to the resulting workload variability. This research proposes a machine learning-based query performance optimization framework that identifies high-priority telemetry records from predicted operating conditions and formalizes a mathematically defined priority-scoring mechanism that a database system can use to allocate indexing, caching, and execution resources adaptively. The framework was developed and validated using the publicly available Environmental Sensor Telemetry Data set (405,184 raw IoT records collected over eight days from three Raspberry Pi-based sensor nodes; cleaned to 385,452 records), with a binary hazard label derived from carbon-monoxide (CO) concentration against an EPA-informed threshold. Seven candidate classifiers, K-Nearest Neighbors, Logistic Regression, an Artificial Neural Network, Gaussian Naive Bayes, Linear Discriminant Analysis, a Support Vector Machine, and CatBoost, were trained on a stratified-representative 10,000-record sample (8,000 train / 2,000 test, class balance within 0.5 percentage points of the full population) and compared on accuracy, precision, recall, F1-measure, prediction variance, and mean absolute error. CatBoost was selected as the proposed model, achieving 99.9% accuracy (95% CI: 99.76–100.00%), compared with 92.1–95.0% for the next-best classifiers. A correlation analysis quantified strong collinearity among CO, LPG, and smoke channels (r 1.00) and a moderate negative association with humidity (r 0.70 to 0.71), which is shown to reduce the effective dimensionality of the sensor feature space and motivates a composite-index recommendation for the underlying time-series store. The trained classifier's output probability is used to define a formal query/index priority score and a resource-allocation objective for adaptive optimization. Results demonstrate that the proposed framework provides a statistically validated, computationally efficient basis for adaptive IoT query optimization; full deployment-scale latency benchmarking against a live database engine is identified as future work.

Item Type:	Article
Status:	Published
DOI:	10.36548/jitdw.2026.4.001
Subjects:	Q Science > Q Science (General) > Q325 Machine learning Q Science > QA Mathematics > QA75 Electronic computers. Computer science Q Science > QA Mathematics > QA76 Computer software Z Bibliography. Library Science. Information Resources > ZA Information resources > ZA4450 Databases
School/Department:	London Campus
URI:	https://ray.yorksj.ac.uk/id/eprint/15833

University Staff: Request a correction | RaY Editors: Update this record