Machine Learning-Based Enhanced Air Quality Estimation with IoT–Cloud Integration and a Mobile Application

Abstract

This paper presents the design and implementation of a real-time Enhanced Air Quality (EAQ) monitoring system integrating Internet of Things (IoT) sensing, cloud computing and supervised machine learning. The system uses an SPS30 particulate sensor and JX‑M electrochemical gas sensors interfaced to a Raspberry Pi 4 (via MCP3008 ADC for analog channels) to continuously measure PM_1.0, PM_2.5, PM₁₀, CO, NO₂, SO₂ and O₃ at one-minute intervals. Raw sensor streams are transmitted to Firebase for cloud storage and processing, where data preprocessing (outlier removal, normalization and missing-value imputation) is applied prior to modelling. Three classifiers—Fuzzy k-Nearest Neighbors (FkNN), Random Forest and Logistic Regression—are trained to predict EAQ classes (Good, Moderate, Poor) and evaluated using k-fold cross-validation with accuracy and F1-score metrics. Experimental results from controlled indoor deployment show that FkNN and Random Forest achieved 99% prediction accuracy, while Logistic Regression attained 98%. A React Native mobile application synchronizes with the Firebase backend to visualize real-time readings, historical trends and EAQ categories. Although the proposed architecture is scalable and low-cost, the current evaluation is limited to indoor conditions; future work will address real-world deployment challenges such as sensor long-term stability and recalibration, network interruptions, power / energy constraints, weather-resistant outdoor operation and external validation against reference-grade monitoring stations.