Enhancing air quality index forecast with string reduction, entropy weight and similarity measure using K-means clustering for fuzzy inference system

This study presents a novel approach to enhance air quality index (AQI) prediction using a fuzzy inference system (FIS) model. A significant challenge in FIS models is the exponential increase in the number of rules with a higher number of inputs, leading to increased computational complexity. To address this issue, this study introduced the string reduction with entropy weight and similarity measure using K-means clustering (SR-EW-SM-KMC) approach. This method enhances model accuracy while reducing computational complexity by leveraging entropy to calculate input weights and scale point factors to determine linguistic values. A similarity measure is then used to form strings, and the elbow method identifies the optimal number of clusters for K-means clustering. The proposed SR-EW-SM-KMC approach is implemented in a FIS model within MATLAB to forecast AQI in both regression and classification scenarios. Statistical analysis is performed against traditional AQI prediction methods and the FIS without rule reduction (FIS-WORR) model. The SR-EW-SM-KMC model's regression performance is validated using metrics: RMSE (0.1342), MSE (0.0180), MAE (0.0955), and MAPE (0.23%) against traditional AQI prediction methods, and RMSE (0.1315), MSE (0.0173), MAE (0.0954), and MAPE (0.29%) against the FIS-WORR model. For classification, its accuracy is assessed using a confusion matrix and ROC analysis, achieving 99% overall accuracy compared to both traditional methods and the FIS-WORR model. Furthermore, the comparative analysis demonstrates that the SR-EW-SM-KMC model significantly outperforms existing models, confirming its effectiveness for accurate and efficient AQI prediction. Additionally, the elbow method's cluster number determination is validated using the silhouette method for splitting the number of clusters, ensuring accurate and efficient clustering. The results demonstrate the SR-EW-SM-KMC approach's superior accuracy and reduced computational complexity, offering a robust solution for AQI prediction. This study emphasises the integration of advanced rule reduction techniques in FIS models and validates the effectiveness of using the silhouette method for optimal cluster determination.