AQ-MultiCal: An Interactive No-Code Machine Learning Platform for Low-Cost Air Quality Sensor Calibration and...

Area of Science:

• Environmental Science
• Data Science
• Sensor Technology

Background:

• Reference-grade air quality monitors are costly, driving the adoption of low-cost sensors (LCS).
• LCS accuracy is compromised by environmental factors, drift, and uncertainty, requiring robust calibration.
• Existing machine learning (ML) calibration methods often require programming expertise, limiting their use.

Purpose of the Study:

• To develop an accessible, no-code platform for calibrating low-cost air quality sensors.
• To standardize and simplify the application of ML models for sensor calibration.
• To enable comparative analysis of various ML calibration models.

Main Methods:

• Introduction of the Air Quality Multi-Model Calibration (AQ-MultiCal) platform.
• Implementation of an interactive, no-code graphical user interface.
• Evaluation of 14 regression models with automated hyperparameter optimization and comparative analysis.
• Validation using CO2 measurements from January-February 2025.

Main Results:

• The AQ-MultiCal platform facilitates the evaluation and comparison of multiple ML calibration models.
• The k-nearest neighbors (kNN) model, optimized via the platform, demonstrated superior performance.
• Achieved a high coefficient of determination (R² = 0.990) with minimal prediction error.

Conclusions:

• AQ-MultiCal provides an effective and accessible solution for accurate low-cost air quality sensor calibration.
• The platform democratizes the use of advanced ML techniques for environmental monitoring.
• Open-source availability enhances accessibility for domain experts without coding skills.