Porting Rulex Software to the Raspberry Pi for Machine Learning Applications on the Edge

Area of Science:

• Computer Science
• Artificial Intelligence
• Edge Computing

Background:

• Edge Computing enables distributed data processing on Internet of Things (IoT) nodes, reducing reliance on central servers.
• Artificial Intelligence (AI) and Machine Learning (ML) are increasingly integral to industrial and preliminary systems.
• Raspberry Pi offers a cost-effective platform for IoT applications.

Purpose of the Study:

• To port the Rulex ML platform onto the Raspberry Pi for Edge Computing applications.
• To evaluate the performance of the ported Rulex platform across diverse datasets and tasks.
• To analyze the power consumption of the Raspberry Pi in an ML-forecasting setup.

Main Methods:

• Porting Rulex ML libraries to Windows 32 Bits, Ubuntu 64 Bits, and Raspbian 32 Bits.
• Conducting ML forecasts on five distinct datasets: Urban Classification, Human Activity detection, Biomedical (mental state), and Vehicle Activity Recognition.
• Implementing image classification using a novel image pre-processing algorithm converting images to time-series data.
• Comparing power consumption between Raspberry Pi and an HP laptop for ML tasks.

Main Results:

• High accuracies achieved in forecasting tasks: 84.13% (Urban Classification), 99.29% (Human Activity - SVM), 95.47% (Vehicle Activity - SVM), and 95.27% (Biomedical - KNN).
• Achieved 96.47% accuracy in edge-based gender classification, outperforming existing literature on varied image datasets.
• Raspberry Pi demonstrated lower energy consumption than an HP laptop for ML tasks, despite longer processing times.

Conclusions:

• The Rulex ML platform is effectively portable to Raspberry Pi for edge-based AI and ML tasks.
• The ported system demonstrates robust performance across various applications, including complex image classification.
• Raspberry Pi offers an energy-efficient solution for deploying ML models at the edge, suitable for IoT environments.