A HYBRID MODEL FOR IMPROVING THE EFFICIENCY OF FOG LAYER IN THE INTERNET OF THINGS

Abstract

Fog computing relies on many interconnected, decentralized devices working together to process fast moving and diverse data. The rapid growth and complexity of this data, in terms of volume, speed, variety, and variability, often slow down data transfer and processing. This delay affects real-time IoT applications like smart healthcare and autonomous vehicles. Despite the importance of responsive data handling, fog layer efficiency has not received enough attention. With more edge devices coming online, there's an urgent need for faster data processing and response. This study developed a hybrid machine learning model combining DBSCAN for clustering, and KNN for classification, to improve fog computing efficiency. A mixed methods approach grounded in pragmatic research philosophy was used. First, a systematic literature review identified key variables influencing fog efficiency. These were ranked using a Random Forest-Recursive Feature Elimination (RF-RFE) algorithm. Input from IoT experts selected purposively was sought through a survey. Snowball sampling was used to identify 30 respondents to validate the ranking. Key variables included Packet Loss Rate, Queue Time, Latency, Channel Utilization, and Throughput. Using a Design Science Research framework, the hybrid DBSCAN+KNN model was developed and tested using the YAFS simulator. The hybrid model outperformed the standard KNN algorithm, achieving 95.33% accuracy, compared to KNN’s 93.67%. It also reached a recall of 99%, surpassing both classic KNN 95.2% and ensemble KNN 89.68% from previous studies. These results show that DBSCAN effectively pre-processes data by grouping similar values and reducing noise before classification by KNN, leading to faster and more accurate processing. This hybrid approach offers a promising way to boost fog efficiency in time-sensitive IoT environments, unlocking the full benefits of smart devices.