AI Analytics Service

The AI Analytics Service is responsible for analyzing environmental sensor data (e.g., temperature, pressure, humidity, wind speed) and predicting PM2.5 air pollution levels using a pre-trained ML model.

---

Purpose

This microservice uses an ML model (trained in Python and exported as .onnx) to:

• Predict air quality (PM2.5 levels)
• Help make proactive environmental decisions
• Support data ingestion and visualization in Grafana

---

Machine Learning Model

• Trained in Python using scikit-learn
• Exported to ONNX format for Java compatibility
• Input: Meteorological data (TEMP, PRES, DEWP, WIND_SPEED, etc.)
• Output: Predicted PM2.5 value (µg/m³)

Example input:

{
  "temp": 20.5,
  "pres": 1015,
  "dewp": 5.2,
  "rain": 0.0,
  "windSpeed": 3.4,
  "wd_NE": true,
  "wd_NW": false,
  "wd_SE": false
}

---

Tech Stack

• Java 21
• Spring Boot 3.4.4
• ONNX Runtime Java
• Prometheus (metrics)
• Kafka (optional future integration)
• Grafana (visualization)

---

Endpoint

Method	Endpoint	Description
POST	/api/ai/predict	Predict PM2.5 from input

---

Sample Request

POST /predict
Content-Type: application/json

{
  "temp": 20.5,
  "pres": 1015,
  "dewp": 5.2,
  "rain": 0.0,
  "windSpeed": 3.4,
  "wd_NE": true,
  "wd_NW": false,
  "wd_SE": false
}

Response:

{
  "prediction": 51.06
}

---

TODOs

MVP Done

• Accept sensor input JSON
• Load .onnx model
• Predict PM2.5 level
• Return prediction via REST

Short-Term Enhancements

• Add Kafka consumer to handle real-time sensor streams
• Log each prediction to a database (PostgreSQL or Redis)
• Add Prometheus custom metrics (e.g., prediction count, average pollution level)
• Add error handling for bad input or model failures
• Return additional metadata in prediction response

Integration Tasks

• Expose predictions to Grafana dashboard
• Send alerts to Notification Service if PM2.5 > threshold
• Secure endpoints via JWT or API Key
• Document API with Swagger/OpenAPI

Advanced Features

• Replace linear regression with LSTM or XGBoost model
• Use model versioning and automatic re-training pipeline
• Create ML pipeline with Airflow or Prefect
• Create a dashboard for visualizing model accuracy
• Integrate with a weather API for real-time external data
• Add batch prediction endpoint (for multiple sensor records)
• Basic monitoring via Prometheus

---

References

• ONNX Runtime Java Docs
• Prometheus + Spring Boot
• Air Quality Dataset Source
• Grafana with Prometheus Guide