An air quality anomaly detection method for data imbalance distribution
By constructing an HPO-LGBM model and combining random forest feature selection and stratified sampling rebalancing methods to optimize model parameters, the problems of overfitting and information loss caused by data imbalance in air quality anomaly detection are solved, achieving more efficient air quality anomaly detection.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- BEIJING UNIV OF TECH
- Filing Date
- 2023-07-27
- Publication Date
- 2026-07-21
AI Technical Summary
Existing air quality anomaly detection models suffer from overfitting and classifiers missing important information when dealing with imbalanced data distributions. Furthermore, existing rebalancing methods are susceptible to noise and have long training times, resulting in poor model performance.
The HPO-LGBM model was adopted, and combined with air quality monitoring station data and abnormal complaint data. The HPO-LGBM model was constructed by random forest feature selection and stratified sampling rebalancing method. The model parameters were optimized by random forest Bayesian hyperparameter optimization algorithm to achieve data rebalancing and model performance improvement.
It improves the accuracy and efficiency of air quality anomaly detection, especially outperforming other models in terms of F1, AUC, G-mean, and MCC values, solving the data imbalance problem and enhancing the model's generalization ability.
Smart Images

Figure CN117076976B_ABST