Sherin Babu

Particulate Matter induced air pollution is known to have significant negative impacts on both the environment and human health. This research evaluates the effectiveness of various decision tree ensemble models in predicting daily PM10 concentrations in Thiruvananthapuram, Kerala, from July 2017 to December 2019. Seven decision tree ensemble models, namely Random Forest, Extra Trees, Gradient Boosting, AdaBoost, LightGBM, XGBoost, and Histogram-Based Gradient Boosting are employed here. To address missing data in the dataset, kNN imputation is utilized for a cohesive dataset suitable for model training. The models utilize both meteorological and air pollutant variables, with performance assessment using metrics such as the coefficient of determination (R²), root mean square error (RMSE) and mean absolute error (MAE). The findings indicate that the Extra Trees regression model provided the best prediction performance (R² = 0.9397, RMSE = 6.664 μg/m³, MAE = 4.950 μg/m³). Histogram-Based Gradient Boosting and Random Forest also demonstrate strong predictive capabilities. The explainability of the best prediction models is conducted by the feature importance analysis process. Feature importance analysis highlighted sulfur dioxide (SO2) as the most significant pollutant influencing PM10 levels, alongside meteorological factors like wind speed and rainfall, enhancing both prediction accuracy and interpretability of results. This research represents the first comprehensive effort to predict PM10 levels in Thiruvananthapuram using machine learning techniques, addressing a gap in regional air quality studies.

Background: One of the most critical contributors to air pollution is particulate matter (PM2.5) that its acute or chronic exposure causes serious health effects to human. Accurate forecasting of PM2.5 concentration is essential for air pollution control and prevention of health complications. A survey of the available scientific literature on random forest model for PM2.5 prediction is presented here. Methods: The scientific literature is extracted from Science Direct database based on a set of specified search criteria. The input features, data length, and evaluation parameters used in PM2.5 prediction were analyzed in this study. Results: The study shows that majority of the publications are aimed at the daily prediction of outdoor PM2.5. Most publications base their PM2.5 prediction on features aerosol optical depth (AOD) and boundary layer height (BLH). PM10 and NO2 are the main air pollutants employed in the PM2.5 estimation. Majority studies utilized input data lengths covering more than one year, and the effectiveness of prediction models are unaffected by the length of investigation. The coefficient of determination, R2 , is the primary evaluation parameter used in all publications. The majority of research study indicated R2 values greater than 0.85, demonstrating the reasonable dependability and efficiency of random forest regression-based PM2.5 prediction models. Conclusion: The study demonstrates that the publications use a variety of meteorological and geological features for PM2.5 estimation, depending on the context of the research as well as data accessibility. The findings demonstrate that it is hard to pinpoint the optimal model in any particular way.

Random forest regression (RFR) is a versatile, easy-to-use and efficient tree based machine-learning algorithm that utilizes the power of multiple decision trees for making decisions. So random forest is a subject of a great deal of research, in many of the machine intelligence applications. The objective of this research is to investigate the scientific output of research based on RFR and to explore its hotspots and frontiers through bibliometric analysis for the years 2007 to 2019. The data are collected from the Web of Science database. The total number of publications, the citations, and types of publications, publication countries, productive authors, prominent journals, and keyword co-occurrence of RFR research are examined, using VOSviewer software. There are 516 papers, published in 299 journals, of which researchers from the USA published 162 articles. The most prolific author, with 6 publications and 240 citations, is Martin H. Teicher. The most cited article is the research article entitled "Genomic selection in wheat breeding using genotyping by sequencing". Among the journals, the most articles (41 publications) are published by the Remote Sensing journal.