Collaborative Filtering to Predict Sensor Array Values in Large IoT Networks.

Area of Science:

• Computer Science
• Data Science
• Sensor Networks

Background:

• Internet of Things (IoT) networks are expanding globally, collecting vast amounts of sensor data.
• Data loss in IoT sensor arrays occurs due to sensor failures, network issues, and environmental factors.
• Uneven sensor precision across different regions can impact data reliability.

Purpose of the Study:

• To propose and evaluate collaborative filtering techniques for predicting missing sensor readings in large-scale IoT networks.
• To leverage existing data within IoT networks to impute lost or erroneous sensor data.
• To compare the effectiveness of different recommender system methods for sensor data imputation.

Main Methods:

• Utilized state-of-the-art recommender system methods, specifically collaborative filtering.
• Tested the proposed approach on two real-world sensor array datasets and one synthetic dataset.
• Conducted experiments by systematically varying the percentage of failed sensors to assess prediction accuracy.

Main Results:

• Achieved a good level of prediction accuracy for missing sensor data.
• Observed that prediction accuracy decreases as the rate of sensor failures increases.
• Identified a failure rate threshold distinguishing the optimal use of memory-based versus model-based approaches.

Conclusions:

• Collaborative filtering is an effective method for predicting missing sensor data in IoT networks.
• The choice between memory-based and model-based methods depends on the sensor failure rate.
• This approach enhances the reliability and utility of data from large-scale IoT deployments.