A Novel Weighted KNN Approach for Predicting Health Impacts Based on Air Quality Data

This paper examines new standards for multilevel classification of air quality-related health impacts. It uses a dataset for air quality and health effects with essential features like pollution concentrations, weather conditions, health impact metrices and Health Impact as a target feature, which classifies health impacts into five levels. As a first step, four baseline classifiers namely Random Forest, Gradient Boost, AdaBoost, and Naive Bayes are applied with the dataset. The baseline models provide a basic understanding of the dataset and its performance for basic classification standards. In the next step, an enhanced k-Nearest Neighbors (KNN) model is proposed to meet the challenge of multi-level classification in air quality and health performance data. A modified distance measure is used to compare the feature importance, improving precision and dynamic neighborhood selection based on data density was applied to improve adaptableness across classes. The experimental results shown that the proposed enhancements to KNN outperformed baseline models with good performance.