Copyright © 2026 Authors retain the copyright of this article. This article is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

@article{207764,
        author = {Joe-Uzuegbu C. K and Emmanuel A. U and Uzoechi L. O and Olubiwe M and Ezigbo P J},
        title = {Integrating Machine Learning Models for Predictive Analysis in Resource-Constrained Environments},
        journal = {International Journal of Innovative Research in Technology},
        year = {2026},
        volume = {13},
        number = {3},
        pages = {2731-2741},
        issn = {2349-6002},
        url = {https://ijirt.org/article?manuscript=207764},
        abstract = {The proliferation of Internet of Things (IoT) deployments and edge computing infrastructures in developing economies has created an urgent demand for predictive analytics that can operate within severe resource constraints. This paper presents a novel lightweight ensemble architecture that integrates quantized decision trees, binarized support vector machines, and prototype-based k-nearest neighbors through an Adaptive Gating Network (AGN) to achieve high-fidelity predictive analysis on microcontrollers with less than 256 KB SRAM and sub-100 MHz processors. We formulate the resource-aware model selection problem as a constrained multi-objective optimization that simultaneously minimizes prediction error, inference latency, and energy consumption. A comprehensive experimental evaluation is conducted on three representative datasets industrial predictive maintenance, agricultural soil monitoring, and urban air quality sensing deployed on an STM32L476 platform. Results demonstrate that the proposed framework achieves 91.4% prediction accuracy while maintaining an inference latency of 12 ms and an energy budget of 2.8 mJ per cycle, outperforming conventional TinyML convolutional networks by 4.2 percentage points in accuracy and reducing energy expenditure by 75.4%. The framework introduces a dynamic voltage and frequency scaling (DVFS) aware gating mechanism that adjusts model complexity at runtime based on available CPU cycles and memory headroom. Extensive ablation studies confirm that adaptive gating contributes a 6.8% accuracy improvement over static ensemble weighting under aggressive resource constraints. The findings establish a practical pathway for deploying robust machine learning models in resource-constrained environments without reliance on cloud offloading, thereby enhancing data privacy, reducing communication overhead, and enabling real-time decision-making in off-grid and low-bandwidth scenarios.},
        keywords = {Resource-constrained machine learning Edge intelligence TinyML Adaptive ensemble methods Predictive analytics Microcontroller deployment},
        month = {August},
        }