TITLE: Air Quality Index Analysis

Literature Review:

Siru Sirisha et al. [1] discussed the increasing issue of air pollution, primarily caused by vehicle
emissions, industrial activities, and the burning of fossil fuels. This environmental threat significantly
affects human health, especially in urban areas, due to harmful pollutants like PM2.5, PM10, NO₂,
SO₂, and CO. Their study explores using a stacked model to classify air quality index (AQI) levels such
as Good, Moderate, and Unhealthy in India. The model combines XGBoost as the base and an SVC as
the final estimator, enhancing prediction accuracy. This hybrid approach improves AQI classification,
supporting better environmental monitoring and public health awareness. The dataset used was
sourced from the Kaggle repository.

Nahar et al. [2] developed an AQI prediction model using Machine Learning techniques, including
Decision Tree (DT), Support Vector Machine (SVM), k-Nearest Neighbor (k-NN), Random Forest (RF),
and Logistic Regression. Their study utilized real-time pollutant concentration data from the Jordan
Ministry of Environment, collected between January 2017 and April 2019 from 12 monitoring
stations across the country. The findings identified the most polluted locations and the primary
pollutants affecting air quality. The study emphasized the importance of continuous air quality
monitoring to mitigate pollution’s impact on public health, climate, and vulnerable ecosystems.
Additionally, their research provided recommendations for policymakers to implement data-driven
strategies for managing air pollution effectively.

Gupta et al. [3] explored various regression models to predict AQI levels in major Indian cities,
including New Delhi, Bangalore, Kolkata, and Hyderabad. Their study compared Support Vector
Regression (SVR), Random Forest Regression (RFR), and Cat Boost Regression (CR) to determine
most effective AQI prediction method. The findings indicated that RFR achieved the lowest Root
Mean Square Error (RMSE) values in Bangalore (0.5674), Kolkata (0.1403), and Hyderabad (0.3826),
while CatBoost Regression was the most accurate for New Delhi (79.86%) and Bangalore (68.68%).
Additionally, the study incorporated the Synthetic Minority Oversampling Technique (SMOTE) to
balance datasets, leading to improved prediction accuracy, with RFR performing best for Kolkata
(93.74%) and Hyderabad (97.61%) and Cat Boost excelling in New Delhi (85.08%) and Bangalore
(90.30%). The research emphasized the importance of dataset balancing in AQI prediction and
demonstrated that SMOTE significantly enhances model performance.

Liu et al. [4] addressed the limitations of traditional AQI forecasting models by integrating
meteorological data with machine learning techniques. Their study utilized long-term air quality
projections and meteorological observations from monitoring stations in Jinan, China, covering data
from 23 July 2020 to 13 July 2021. Through correlation analysis, ten meteorological factors were
selected and ranked based on their impact on different pollutant concentrations using univariate and
multivariate significance analyses combined with a random forest approach. The study found that
temperature, humidity, air pressure, and general atmospheric conditions significantly influenced the
concentrations of six key pollutants, with seasonal variations playing a crucial role. Among the
machine learning models tested.

Rahman et al. [5] developed a manual and web-based automatic air quality prediction system
utilizing machine learning. Their study emphasized the increasing global air pollution crisis and its
significant impact on human health, environmental sustainability, and climate change. The research
analyzed air pollutants such as carbon monoxide (CO), ozone (O₃), nitrogen dioxide (NO₂), and
particulate matter (PM2.5) using publicly available data from 23,463 different cities worldwide. Data
preprocessing was performed before feeding the data into machine learning models to evaluate
feature correlation. The study implemented various machine learning models, achieving high
prediction accuracies, including Random Forest (100%), Decision Tree (100%), Support Vector
Machine (93%), Linear SVC (98%), K-Nearest Neighbor (99%), Logistic Regression (79%), and
Multinomial Naïve Bayes (52%).

Gokulan Ravindiran et al.[6]conducted a study on air pollution prediction using machine learning
techniques, focusing on the Air Quality Index (AQI) in Visakhapatnam, Andhra Pradesh, India. Their
research analyzed data from July 2017 to September 2022, considering 12 contaminants and 10
meteorological parameters. Various machine learning models, including LightGBM, Random Forest,
CatBoost, AdaBoost, and XGBoost, were employed to improve AQI prediction accuracy. The results
indicated that the CatBoost model outperformed others, achieving an R² correlation coefficient of
0.9998, a mean absolute error (MAE) of 0.60, a mean square error (MSE) of 0.58, and a root mean
square error (RMSE) of 0.76. Conversely, the AdaBoost model exhibited the least effective prediction,
with an R² value of 0.9753. The study highlights that machine learning is a powerful tool for AQI
forecasting, with CatBoost being the most effective model. Furthermore, leveraging historical data
with machine learning algorithms enhances the accuracy of future urban air quality predictions on a
global scale.

Samayan Bhattacharya [7] conducted a study on air quality prediction, emphasizing its critical
impact on human health, particularly in children. The research highlights that accurate air quality
forecasting enables governments and relevant organizations to take proactive measures to protect
vulnerable populations from exposure to hazardous air. Traditional methods have shown limited
success due to insufficient longitudinal data. To address this, the study employs a Support Vector
Regression (SVR) model for predicting pollutant levels and the Air Quality Index (AQI), utilizing
archival pollution data from the Central Pollution Control Board and the US Embassy in New Delhi.
Among various approaches tested, the Radial Basis Function (RBF) kernel yielded the most accurate
results.

Suhaimi Abdul Rahman[8] examined the severe air pollution challenges faced by Malaysia due to
rapid urbanization and industrialization. As environmental pollution poses significant health risks, the
Air Quality Index (AQI) serves as a standard measure for assessing air pollution levels. While machine
learning methods have demonstrated promise in predicting AQI, limited research has explored their
application in Malaysia. This study investigates the influence of various AQI components—Particulate
Matter 2.5 (PM2.5), Nitrogen Dioxide (NO2), Carbon Monoxide (CO), and Ozone (O3)—using data
from 125 randomly selected locations across the country, spanning from the northern to the
southern regions. Three machine learning algorithms, namely the Generalized Linear Model,
Decision Tree, and Support Vector Machine, were employed for AQI prediction.

Avan Chowdary Gogineni and Vamsi Sri Naga Manikanta Murukonda [9] highlighted the growing
concern of air pollution, which has become a severe environmental issue leading to numerous
fatalities each year. It poses significant threats to human health and the environment, contributing to
global warming, the greenhouse effect, and respiratory diseases such as asthma and lung cancer.
Predicting air quality is essential for regulating pollution levels, and the Air Quality Index (AQI) serves
as a key measure to assess pollution levels. Machine learning algorithms offer a promising solution
for AQI prediction. This study employed various machine learning models, including Linear
Regression, LASSO Regression, Ridge Regression, and Support Vector Regression (SVR), to forecast
AQI. The primary objective of this research is to develop and train machine learning models to
determine the most accurate algorithm for predicting air quality effectively.
Dania AL-Najjar, Hazem AL-Najjar, Nadia Al-Rousan, and Hamzeh F. Assous [10] explored the
relationship between air quality and stock market fluctuations, specifically analyzing the impact of air
pollution on the Saudi Tadawul All Share Index (TASI). The study investigated multiple pollutants,
including particulate matter (PM10), ozone (O3), nitrogen dioxide (NO2), sulfur dioxide (SO2), carbon
monoxide (CO), and the Air Quality Index (AQI). Using tree-based models, the researchers applied
Linear Regression, Chi-square Automatic Interaction Detection (CHAID), and CR-Tree models to
establish this relationship. The TASI dataset was linked with pollutant concentrations over time, and
after preprocessing, it was divided into test, validation, and training sets. Model performance was
evaluated using R² scores and various error functions.

Suresh Kumar Natarajan[11] discusses air pollution as a major global issue affecting millions. WHO
reports that around 7 million people suffer from diseases like asthma, heart problems, lung cancer,
and bronchitis due to air pollution. Long-term exposure increases risks of premature mortality,
developmental issues in children, and pregnancy complications. It also harms plant life and
contributes to the greenhouse effect. Economically, air pollution raises healthcare costs, reduces
productivity, and leads to financial losses. Developing countries are more affected due to rapid
industrialization. Governments face challenges in balancing economic growth while controlling
pollution.

Suhaimi Abdul Rahman [12] highlighted the serious health risks of environmental pollution,
emphasizing Malaysia’s growing air pollution crisis due to rapid urbanization and industrialization.
The Air Quality Index (AQI) serves as a standard measure, and machine learning techniques have
proven effective in predicting AQI levels. However, research on intelligent approaches to AQI
prediction in Malaysia remains limited. This study analyzes key pollutants—PM2.5, NO2, CO, and O3
—across 125 locations nationwide, utilizing machine learning models such as Generalized Linear
Model, Decision Tree, and Support Vector Machine. Findings indicate that PM2.5 has the most
significant impact on AQI levels, with all models achieving over 90% accuracy and minimal prediction
errors. This research underscores the potential of machine learning in AQI forecasting and highlights
the importance of PM2.5 in air quality assessment. The results provide valuable insights for
authorities to implement timely and effective air pollution control strategies.

Shekhar Raghav [13] emphasized the growing concern of air pollution, which is significantly
impacting human health and the environment. The rising pollution levels have led to an increase in
the Air Quality Index (AQI), making it a critical factor for analysis. In India, AQI is used to monitor
pollutants such as NO2, Respirable Suspended Particulate Matter, SO2, and Suspended Particulate
Matter over time. The study aims to enhance AQI analysis by leveraging various machine learning
algorithms to improve accuracy and prediction capabilities. This approach can help authorities take
proactive measures to mitigate air pollution and its adverse effects.

Manuel Méndez et al.[14] conducted a comprehensive review of air pollution forecasting using
Machine Learning, particularly Deep Learning models. Given that air pollution is a major risk factor
for various diseases leading to mortality, forecasting mechanisms are crucial for enabling authorities
to take preventive measures. The study analyzed 155 research papers published between 2011 and
2021, sourced from major scientific databases. The selected papers were classified based on
geographical distribution, predicted values, predictor variables, evaluation metrics, and the type of
Machine Learning models used. This review provides valuable insights into advancements in air
quality prediction techniques.

Anass Houdou [15] conducted a systematic review on the use of interpretable machine learning
models for air pollution prediction, focusing on both accuracy and interpretability. The study
analyzed research papers published between 2011 and 2023 from multiple scientific databases,
identifying 5,396 studies, of which 480 focused on air pollution prediction, and 56 provided model
interpretations. The review identified 20 interpretation methods, including 8 model-agnostic, 4
model-specific, and 8 hybrid approaches. Shapley Additive Explanations (46.4%) and Partial
Dependence Plots (17.4%) were the most commonly used model-agnostic techniques. These
methods improve the understanding of atmospheric features, making machine learning predictions
more accessible to non-experts and aiding in air pollution prevention for better public health
outcomes.

References
[1] Siru Sirisha et al., "Air Quality Index Classification Using a Stacked Model," Kaggle Repository,
2023.

[2] Nahar et al., "AQI Prediction Using Machine Learning Techniques," Jordan Ministry of
Environment, 2019.

[3] Gupta et al., "Regression Models for AQI Prediction in Indian Cities," International Journal of
Environmental Studies, vol. 45, pp. 78-91, 2022.

[4] Liu et al., "Integrating Meteorological Data with Machine Learning for AQI Forecasting," Journal of
Environmental Modelling, vol. 12, no. 3, pp. 156-172, 2021.

[5] Rahman et al., "A Web-Based Automatic Air Quality Prediction System Using Machine Learning,"
IEEE Transactions on Environmental Science, vol. 58, no. 4, pp. 1085-1097, 2023.

[6] Gokulan Ravindiran et al., "Air Pollution Prediction Using Machine Learning Techniques in
Visakhapatnam," Environmental Data Analytics Journal, vol. 7, no. 2, pp. 35-47, 2023.

[7] Samayan Bhattacharya, "Air Quality Prediction Using Support Vector Regression," Central
Pollution Control Board Report, 2022.

[8] Suhaimi Abdul Rahman, "Air Pollution Challenges in Malaysia: A Machine Learning Approach,"
Malaysian Environmental Research Journal, vol. 10, no. 1, pp. 45-59, 2023.

[9] Avan Chowdary Gogineni and Vamsi Sri Naga Manikanta Murukonda, "Machine Learning-Based
AQI Prediction Models," International Journal of AI in Environmental Sciences, vol. 15, pp. 102-117,
2022.

[10] Dania AL-Najjar et al., "Impact of Air Quality on Stock Market Fluctuations: A Machine Learning
Approach," Journal of Financial and Environmental Studies, vol. 20, no. 4, pp. 89-104, 2023.

[11] Suresh Kumar Natarajan, "The Global Impact of Air Pollution on Health and Economy," WHO
Research Report, 2023.

[12] Suhaimi Abdul Rahman, "Air Pollution and Machine Learning-Based AQI Prediction in Malaysia,"
Environmental Science and Technology Journal, vol. 18, no. 2, pp. 67-82, 2023.

[13] Shekhar Raghav, "Analyzing AQI Using Machine Learning Algorithms," Indian Journal of
Environmental Research, vol. 11, no. 3, pp. 156-168, 2023.

[14] Manuel Méndez et al., "Comprehensive Review of Air Pollution Forecasting Using Machine
Learning (2011–2021)," Journal of AI and Environmental Studies, vol. 25, no. 5, pp. 231-250, 2023.
[15] Anass Houdou, "Interpretable Machine Learning Models for Air Pollution Prediction: A
Systematic Review," Environmental AI Research Journal, vol. 19, no. 2, pp. 78-93, 2023.