Litcius/Paper detail

Data Driven Multivariate Air Quality Forecasting using Dynamic Fine Tuning Autoencoder Layer

K. Krishna Rani Samal, Korra Sathya Babu, Ankit Kumar Panda, Santos Kumar Das

202011 citationsDOI

Abstract

Particulate matter 2.5 (PM2.5) has a severe negative impact on human health, so forecasting of particulate matter has become one of the primary concerns worldwide as it has a significant role in effective control of air pollution. In the current era, the application of artificial intelligence techniques is growing fast, improving environmental air quality. So this research work introduced a data-driven SVR-Autoencoder model for trend analysis and dynamic data modeling in the real-time environment. It also introduced Multivariate imputation by chained equations (MICE) as a multiple imputation method, which considers meteorological attributes to impute the missing values of the PM2.5 dataset. We further added LSTM as a Stacked Denoising Autoencoder layer, which dynamically adjusts the weight to provide accurate forecasting results in a real-time environment. The experimental result shows that the SVR-Autoencoder model provides 34-67% better results than the baseline models, which indicates its effectiveness in air quality modeling.

Topics & Concepts

AutoencoderMultivariate statisticsImputation (statistics)Computer scienceAir quality indexMissing dataArtificial intelligenceData miningData modelingBaseline (sea)Machine learningParticulatesDeep learningMeteorologyEcologyDatabaseGeologyBiologyPhysicsOceanographyAir Quality Monitoring and ForecastingAir Quality and Health ImpactsEnergy Load and Power Forecasting