Scholarlink

<p>The proliferation of Internet of Things (IoT) devices and smart edge systems has intensified the need for real-time predictive analytics directly on resource-constrained hardware. Traditional cloud-based approaches often fail to meet stringent latency and bandwidth requirements, motivating the adoption of lightweight on-device intelligence. This paper presents a low-weight deep learning architecture that has been optimized to run on edge computing systems due to small network architecture, model compression, and low-cost inference techniques. The framework is deployed and evaluated on a Raspberry&#xa0;Pi&#xa0;4 using TensorFlow Lite, demonstrating strong performance across predictive and computational metrics. Experimental results show improved accuracy (0.9337) and significantly reduced latency (12.74&#xa0;ms) compared with state-of-the-art models such as MobileNetV3 and MCUNet. Additional analysis reveals high throughput, low energy consumption, and robust class-wise prediction stability, confirmed through confusion matrix and statistical significance testing. Overall, the proposed framework offers an effective balance between accuracy and efficiency, making it well-suited for real-time edge-based predictive analytics applications.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A Lightweight Deep Learning Framework for Real-Time Predictive Analytics in Edge Computing Environments

  • 期刊论文

  • 发表日期：

    2026年4月13日

查看全文
  • Nirmalkumar S. Benni,
  • B. C. Anil,
  • S. R. Jayasimha,
  • K. Girish,
  • C. Vishal,
  • Samitha Khaiyum

摘要

The proliferation of Internet of Things (IoT) devices and smart edge systems has intensified the need for real-time predictive analytics directly on resource-constrained hardware. Traditional cloud-based approaches often fail to meet stringent latency and bandwidth requirements, motivating the adoption of lightweight on-device intelligence. This paper presents a low-weight deep learning architecture that has been optimized to run on edge computing systems due to small network architecture, model compression, and low-cost inference techniques. The framework is deployed and evaluated on a Raspberry Pi 4 using TensorFlow Lite, demonstrating strong performance across predictive and computational metrics. Experimental results show improved accuracy (0.9337) and significantly reduced latency (12.74 ms) compared with state-of-the-art models such as MobileNetV3 and MCUNet. Additional analysis reveals high throughput, low energy consumption, and robust class-wise prediction stability, confirmed through confusion matrix and statistical significance testing. Overall, the proposed framework offers an effective balance between accuracy and efficiency, making it well-suited for real-time edge-based predictive analytics applications.