THROUGHPUT PREDICTION FRAMEWORK ON MACHINE LEARNING-BASED WIRELESS SENSOR NETWORKS AND M- IOT

• Main Author: Eliviani, Rosa
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/70813
• Institution: Institut Teknologi Bandung

The Internet of Things (IoT) is a developmental concept where sensors and intelligence are added to hardware items and have links with the internet. The characteristics of using IoT support sending data in multimedia form. However, this poses a problem because multimedia data creates a large bandwidth and is sensitive to delays. Several previous studies have proposed M-IoT architectures to analyze, process, and use resources more reliably. Apart from that, another problem with Multimedia IoT lies in its multimedia traffic. This problem is also related to problems in the Wireless Sensor Networks (WSN), namely problems in sending high-quality data, while the devices used are devices with limited resources. Based on this, it is necessary to have a solution to improve the problem of sending data by improving its quality so that it can improve the quality of the Quality of Experience (QoE). To achieve the goal of improving data transmission, it is necessary to have good network management by paying attention to network speed, namely throughput. So it is necessary to analyze and predict throughput for network management. A good throughput prediction can measure QoE significantly, where this QoE shows the ability of a network to provide good service. Therefore, this study analyzes how to predict the throughput of Wireless Sensor Networks. First, what has been done is a comparative study with throughput predictions in Multimedia IoT. Next, it undertook to develop a machine learning-based predictive framework. The Throughput Prediction Framework identifies the most important characteristics and uses these characteristics to predict real-time throughput. The final stage is to experiment with the framework. The evaluation shows that the WSN-IoT prediction is quite good. For a time, 1-second breakdown, the Mean Absolute Percentage Error (MAPE) for all investigated scenarios is in the range of one to 8 percent.