Al- Based Air Quality Monitoring

A. Background and Context Air quality is influenced by transport emissions, industrial activity, construction dust, fuel combustion, weather, and local geography. Particulate matter, particularly PM2.5, is important because fine particles can penetrate deeply into the respiratory system. The World Health Organization identifies ambient air pollution as a major health risk and provides evidence-based air-quality guideline values [1]. Monitoring data can help communities, researchers, and local authorities understand pollution patterns, but observations need to be timely, spatially representative, and communicated clearly. Low-cost sensors and connected microcontrollers make it possible to deploy supplementary monitoring nodes in locations where dense coverage is useful. Artificial intelligence can help identify unusual readings, estimate missing values, classify pollution conditions, and forecast short-term trends. However, low-cost devices are not automatically equivalent to regulatory monitoring stations: sensor response, humidity, temperature, drift, and installation conditions can affect readings. B. Problem Statement A practical monitoring solution must do more than display a sensor number. It should preserve timestamps and units, flag invalid or stale observations, distinguish measured data from estimates, provide useful visual trends, and communicate uncertainty. A poorly calibrated sensor or an unvalidated model can create misleading warnings. The problem addressed in this paper is the design of a cost-conscious, connected monitoring framework with explicit data-quality checks and a responsible path toward AI-assisted analysis.

C. Aim and Objectives

• Design a modular IoT architecture for collecting local air-quality and environmental observations.
• Specify validation, calibration, storage, and dashboard processes for time-series data.
• Define a transparent baseline for status classification and identify where AI can support anomaly detection and forecasting.
• Propose measurable evaluation methods without fabricating experimental results.
• Identify reliability, privacy, security, and deployment considerations for a student-scale prototype. D. Scope and Contribution The proposed work focuses on supplementary outdoor or semi-outdoor monitoring using a sensor node, backend service, database, and dashboard. Its main contribution is an end-to-end design that treats sensor quality and uncertainty as first-class requirements alongside AI analysis. The paper is a literature-informed proposal, not a report of a completed field trial.

A. Public-Health and Monitoring Context WHO guidance highlights the health relevance of particulate matter and other pollutants and recommends evidence-based approaches to air-quality management [1]. Reference-grade monitoring remains important for official reporting, while supplementary sensor networks can provide additional spatial and temporal detail when carefully calibrated and interpreted. B. IoT and Machine Learning for Monitoring Gryech et al. reviewed machine-learning and IoT approaches for outdoor air-pollution monitoring and prediction, grouping approaches into time-series, feature-based, and spatiotemporal methods. Their review also identified recurring challenges involving limited coverage, data diversity, and context-specific features [2]. These findings support a modular system that records environmental context and preserves location and time metadata rather than relying only on a single pollutant value. C. AI for Air-Quality Prediction Recent surveys describe deep-learning approaches for modelling temporal and spatial patterns in air-quality data, while emphasizing the importance of dataset selection and prediction design [3]. A 2025 systematic review of AI in air-pollution monitoring and forecasting reports applications in data completion, sensor calibration, anomaly detection, AQI estimation, and short-term forecasting [4]. These application areas map naturally to separate modules, but each model requires a suitable dataset and evaluation procedure. D. Identified Research Gap Many prototype systems emphasize live display, whereas prediction studies may focus on model metrics without explaining how sensor faults or data gaps are handled in deployment. A useful student-scale design should combine low-cost sensing with explicit data-quality flags, clear separation of observed and predicted values, fail-safe behaviour during connectivity loss, and a repeatable evaluation plan. This proposal addresses that integration gap at the design level. Research theme Design implication Low-cost sensing Treat readings as indicative until calibrated against a suitable reference instrument. IoT data pipelines Attach timestamp, device ID, units, sensor status, and firmware version to each record. AI prediction Compare models against a simple baseline and use time-ordered validation. User communication Label estimates and forecasts separately from measured observations; explain alert thresholds.

A. System Architecture The proposed architecture contains five logical stages: sensing, edge processing, secure communication, backend storage and analysis, and user-facing presentation. A sensor node measures selected pollutant indicators and environmental conditions. An ESP32-class controller timestamps and validates the measurements, then transmits records through Wi-Fi using HTTPS or MQTT with authentication. A backend API validates incoming payloads, stores observations, computes visual summaries, and invokes trained models where available. The dashboard presents recent readings, historical charts, data-quality flags, and alerts.

A. Limitations

• Low-cost sensors can drift and may respond to humidity, temperature, or interfering gases.
• A single node measures local conditions and cannot represent an entire city or neighbourhood.
• Official AQI calculations require the correct pollutant inputs and jurisdiction-specific rules.
• Forecast models require sufficient labelled data and may not generalize across sites or seasons.
• Connectivity, power, enclosure quality, maintenance, and calibration affect long-term operation.
• This paper proposes a design and evaluation protocol; it does not report field-validated performance. B. Responsible Use The system should communicate uncertainty, data freshness, units, and limitations in plain language. Alerts should not imply medical diagnosis or official regulatory status unless formally authorized. If the data are shared publicly, location precision and retention should be chosen carefully. Model changes should be versioned and reviewed, and critical decisions should not depend on an unvalidated prediction alone. C. Future Scope Future work can add weather-service integration, multi-node spatial mapping, edge inference, explainable anomaly detection, uncertainty intervals, automated sensor-drift checks, and energy-efficient solar power. A larger study could compare multiple sensor models against reference stations, investigate seasonal transferability, and assess whether forecasts improve public communication or local environmental planning.

This paper presents a literature-informed framework for AI-based air quality monitoring that integrates connected sensing, data validation, secure storage, dashboard visualization, and staged machine-learning analytics. The architecture emphasizes calibration, explicit data-quality flags, baseline comparisons, and a clear distinction between measured observations, calculated indicators, and forecasts. The proposal offers a practical path for a student-scale prototype while recognizing that low-cost sensors and AI models require rigorous validation before high-stakes use. Future implementation should prioritize reference-based testing, dependable data collection, and transparent communication of uncertainty.