研究目的
Investigating the effectiveness of machine learning algorithms for improved data analysis of biological aerosol using the WIBS.
研究成果
Gradient boosting offered the best performance consistently across different data preparation strategies and data sets. DBSCAN was identified as a potential alternative to HAC in the absence of laboratory training data, though its effectiveness varies. The study underscores the importance of collecting additional laboratory generated aerosol to enhance the analysis of ambient data and the potential of higher spectral instruments for more accurate classification.
研究不足
The study highlights the dependency of HAC on data preparation and the cluster index used, with inconsistent results across different sets of laboratory generated aerosol. DBSCAN's effectiveness is also dependent on data preparation and parameter selection, with varying performance across data sets. The need for additional laboratory generated aerosol to improve interpretation of current databases is emphasized.
