ADAPTIVE REAL-TIME CLEANING OF HETEROGENEOUS SENSOR DATA IN SMART HOME SYSTEMS BASED ON NOISE CLASSIFICATION

Abstract

Modern smart home systems generate substantial volumes of data, the quality of which is critically important for effective management, analysis, and forecasting. However, raw data streams often contain complex combinations of noise and anomalies, including outliers, concept drift, and periods of stagnant values. Such artifacts significantly reduce data reliability, potentially leading to incorrect operation of intelligent systems and flawed decision-making. Existing data cleaning methods often demonstrate limited effectiveness when dealing with heterogeneous types of noise in real-time, particularly when processing sensor data with varying physical characteristics.

This paper presents a novel method — ACRA (Adaptive Classification-based Real-time Anomaly cleaning) — designed for adaptive cleaning of heterogeneous sensor time series data in the context of smart homes. The ACRA method incorporates a classifier based on an ensemble of decision trees (Random Forest) to identify specific types of noise within a sliding window of incoming data. The classification results are supplemented by a heuristic rule that leverages window variance analysis to more accurately detect low-noise periods. Based on the combination of the classifier’s output and the rule, an adaptive strategy module is activated to dynamically select the most appropriate operator for correcting the current data value. To evaluate the effectiveness of the proposed method, an experimental study was conducted using real-world time series of temperature, humidity, and total energy consumption collected in a residential environment, with synthetically injected controlled noise types.

The experimental results demonstrated that the ACRA method provides significantly better cleaning quality compared to common baseline methods such as the moving median and Kalman filter across all three types of sensor data studied. The proposed ACRA method is a robust and flexible tool for enhancing the quality of sensor data in smart home systems, laying the foundation for the development of more accurate and efficient intelligent applications.