Scholarlink

VITALAIR is an innovative AI-based forecasting system designed to predict both the Air Quality Index (AQI) with exceptional accuracy. By harnessing advanced deep learning techniques together with probabilistic regression, the platform effectively addresses the challenges associated with modeling complex and nonlinear environmental datasets. In this study, we propose an ensemble framework that combines the N-BEATS (Neural Basis Expansion Analysis for Time Series) model with Gaussian Process Regression (GPR) to improve prediction accuracy and provide reliable uncertainty estimates. Real-time pollutant data collected from CPCB monitoring stations is processed through a robust data pipeline, ultimately yielding 28-day forecasts. Evaluation based on metrics such as Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), Root Mean Squared Error (RMSE), and the coefficient of determination (R2) confirms that the ensemble method significantly outperforms conventional single-model approaches. The scalable design of VITALAIR supports forecasting across multiple regions and offers valuable insights for urban planners, policymakers, and public health authorities. This paper details the development, evaluation, and potential of combining deep learning with probabilistic methods for enhanced environmental risk assessment and sustainable urban management.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Forecasting of Air Quality Index Based on Pollutant Concentrations Using Gaussian Process Regression and N-BEATS Model

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Yogesh Sajithkumar,
  • Rasith Novfal,
  • Sabiyath Fatima

摘要

VITALAIR is an innovative AI-based forecasting system designed to predict both the Air Quality Index (AQI) with exceptional accuracy. By harnessing advanced deep learning techniques together with probabilistic regression, the platform effectively addresses the challenges associated with modeling complex and nonlinear environmental datasets. In this study, we propose an ensemble framework that combines the N-BEATS (Neural Basis Expansion Analysis for Time Series) model with Gaussian Process Regression (GPR) to improve prediction accuracy and provide reliable uncertainty estimates. Real-time pollutant data collected from CPCB monitoring stations is processed through a robust data pipeline, ultimately yielding 28-day forecasts. Evaluation based on metrics such as Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), Root Mean Squared Error (RMSE), and the coefficient of determination (R2) confirms that the ensemble method significantly outperforms conventional single-model approaches. The scalable design of VITALAIR supports forecasting across multiple regions and offers valuable insights for urban planners, policymakers, and public health authorities. This paper details the development, evaluation, and potential of combining deep learning with probabilistic methods for enhanced environmental risk assessment and sustainable urban management.