Regression based hybrid machine learning model performance evaluation on air quality index prediction in Kolkata

Description

Muni Lakshmi G K and Mokesh Rayalu G build in 2026 machine learning forecasts of the Air Quality Index for Kolkata, India, where fine particle pollution threatens public health. Their dataset blended weather variables and pollutant readings, cleaned through a multi-step method for filling missing values. Many individual algorithms were tried, from linear regression and decision trees to neural networks and a sequence model. Combined ensemble methods that average or stack several learners were tried too, alongside simple baselines for reference.

A voting ensemble achieved the best overall error and fit scores in the temporal test split, while a stacking ensemble led on two other error scores in the reported results. Sequence modelling captured time patterns yet trailed the ensembles. Removing the two particle readings degraded every model clearly, and checks confirmed those particles as the dominant predictors, with stability checks supporting the framework.

Catalog record

Source: https://europepmc.org/article/MED/PMC13500472