Deep-Reinforcement-Learning-Based IoT Sensor Data Cleaning Framework for Enhanced Data Analytics.

Area of Science:

• Computer Science
• Artificial Intelligence
• Data Science

Background:

• Internet of Things (IoT) generates vast amounts of sensor data crucial for smart city applications.
• Noisy, redundant, or empty IoT data negatively impacts machine learning and deep learning algorithm performance.
• Effective data cleaning is essential to improve the reliability of intelligent IoT systems.

Purpose of the Study:

• To propose a novel deep reinforcement learning (deep RL) framework for cleaning Internet of Things (IoT) sensor data.
• To address the challenge of noisy, redundant, and empty data in IoT sensor streams.
• To enhance the performance of smart city applications by improving data quality.

Main Methods:

• A deep reinforcement learning (deep RL) framework utilizing a deep Q-network (DQN) agent.
• The DQN agent classifies sensor data into 'empty', 'garbage', or 'normal' categories.
• Input features include current and past Received Signal Strength (RSS) values; rewards are based on predicted actions.

Main Results:

• The proposed deep RL system achieved approximately 96% accuracy in data cleaning after the exploration phase.
• Experimental results show superior performance compared to a traditional time-series-based fully connected neural network (FCDQN) approach.
• The system effectively identifies and removes irrelevant and harmful data points from sensor streams.

Conclusions:

• Deep reinforcement learning offers a powerful approach for robust IoT sensor data cleaning.
• Improved data quality through deep RL significantly enhances the performance of intelligent IoT applications.
• This framework has the potential to increase the reliability and efficiency of smart city initiatives.