Scholarlink

The Internet of Things (IoT) makes it easier for the physical and virtual worlds to connect and interact, which generates enormous amounts of data, or “Big Data.” This chapter explores several machine learning techniques that address IoT data processing challenges, with a focus on smart city applications. The study provides an overview of the methods used to extract higher-level information from the data, along with a taxonomy of machine learning algorithms. Every model is trained on the dataset as part of the research, and the accuracy and training time are recorded. The findings indicate that Stochastic Gradient Descent attains the maximum accuracy while Mini-Batch Gradient Descent performs better in terms of training time. The model’s performance on the test data is indicated by the accuracy that BGD is able to attain. An accuracy of 0.7733 means the model correctly predicts 77.33 instances. Despite the longer training time, SGD achieves a higher accuracy of 0.8300 (83.00%). This can be attributed to its ability to escape local minima more effectively due to the noisy updates. MBGD achieves the highest accuracy of 0.8567 (85.67%) while maintaining stable convergence, leading to better generalization of the test data. This analysis offers insightful information on how well different techniques handle and evaluate IoT data for applications in smart cities.

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

Improving Machine Learning Models and Neural Network Performance in IoT Systems Using Gradient Descent Methods

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • N. Rakesh,
  • G. L. Prakash,
  • Mohamed Shakir

摘要

The Internet of Things (IoT) makes it easier for the physical and virtual worlds to connect and interact, which generates enormous amounts of data, or “Big Data.” This chapter explores several machine learning techniques that address IoT data processing challenges, with a focus on smart city applications. The study provides an overview of the methods used to extract higher-level information from the data, along with a taxonomy of machine learning algorithms. Every model is trained on the dataset as part of the research, and the accuracy and training time are recorded. The findings indicate that Stochastic Gradient Descent attains the maximum accuracy while Mini-Batch Gradient Descent performs better in terms of training time. The model’s performance on the test data is indicated by the accuracy that BGD is able to attain. An accuracy of 0.7733 means the model correctly predicts 77.33 instances. Despite the longer training time, SGD achieves a higher accuracy of 0.8300 (83.00%). This can be attributed to its ability to escape local minima more effectively due to the noisy updates. MBGD achieves the highest accuracy of 0.8567 (85.67%) while maintaining stable convergence, leading to better generalization of the test data. This analysis offers insightful information on how well different techniques handle and evaluate IoT data for applications in smart cities.