Ensemble Transfer Learning for Air Quality Classification: A Robust

Model for Environmental Monitoring

Venu K, Krishnakumar B, Sasipriyaa N,

Deepak Raajan N, Dharaneesh S and Hari R P

Kongu Engineering College, Erode, Tamil Nadu, India

Keywords: Air Pollution Prediction, Transfer Learning, Ensemble Model, VGG16, ResNet, EfficientNet.

Abstract: One of the most urgent environmental issues facing the world today is air pollution, which has an immediate

impact on ecosystems and human health. Predicting air quality accurately is crucial for mitigation plans and

early warning systems. The classification of air quality into six predetermined categories—Good, Moderate,

Unhealthy for Sensitive Groups, Unhealthy, Very Unhealthy, and Hazardous/Severe—is examined in this

paper using transfer learning techniques. A dataset of images representing air quality levels was used, with

pre-trained models such as VGG16, ResNet, and InceptionV3 fine-tuned for this classification task. Transfer

learning models, known for their efficiency in image-based tasks, were individually tested and compared

based on classification accuracy and performance metrics. To further improve prediction accuracy, a novel

ensemble approach was implemented, combining VGG16, ResNet, and EfficientNet into a unified model.

The ensemble model achieved significantly higher accuracy compared to individual models, particularly in

predicting complex air quality scenarios such as the Hazardous/Severe category. This improvement in

accuracy underscores the potential of combining multiple pre-trained models in air quality prediction tasks,

addressing the challenge of differentiating between closely related pollution levels. The results suggest that

this ensemble approach not only enhances classification accuracy but also provides a more robust prediction

framework for real-world applications. The proposed method shows promise for integration into real-time air

quality monitoring systems, offering an effective tool for public health agencies to predict and respond to

deteriorating air quality conditions. This study adds to the expanding corpus of research on transfer learning

and how it's used in environmental monitoring.

1 INTRODUCTION

One of the biggest threats to the environment and

public health in the world today is air pollution. The

World Health Organization (WHO) estimates that air

pollution causes millions of premature deaths

annually, mostly from cardiovascular and respiratory

conditions that are made worse by extended exposure

to dangerous air pollutants. These pollutants, which

all contribute to the deterioration of air quality,

include carbon monoxide (CO), nitrogen dioxide

(NO₂), sulfur dioxide (SO₂), particulate matter (PM),

and ozone (O₃). To lessen the negative consequences

of poor air quality and enable prompt warnings and

the implementation of preventive actions, accurate air

pollution prediction systems are crucial. Due to their

capacity to handle large volumes of data and generate

accurate forecasts, machine learning and deep

learning approaches have been increasingly popular

in recent years for the prediction of air pollution.

Traditional air pollution prediction models, such as

chemical transport models or statistical regression

methods, often rely on handcrafted features and a

deep understanding of atmospheric dynamics. While

these models are effective, they face challenges in

accurately predicting complex air quality patterns,

especially in environments with rapid fluctuations in

pollution levels. Deep learning algorithms have

therefore become a viable substitute because of their

capacity to automatically extract high-level

characteristics from big datasets without the need for

manual intervention.

Convolutional Neural Networks (CNNs), one of

the deep learning methods, have shown exceptional

efficacy in image categorization tasks. CNNs are

appropriate for predicting air pollution levels from

sensor images or data visualizations because they can

learn to identify intricate patterns in data. However,

K, V., B, K., N, S., N, D. R., S, D. and R P, H.

Ensemble Transfer Learning for Air Quality Classiﬁcation: A Robust Model for Environmental Monitoring.

DOI: 10.5220/0013609300004664

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 3rd International Conference on Futuristic Technology (INCOFT 2025) - Volume 3, pages 91-98

ISBN: 978-989-758-763-4

building deep neural networks from the ground up

frequently calls for substantial computational

resources and huge datasets. By allowing the use of

pre-trained models that have been refined on huge

picture datasets like ImageNet, transfer learning

provides a workable solution to this issue. High

accuracy can be attained without a large dataset by

optimizing these pre-trained models for the particular

goal of air pollution prediction.

In addition to testing individual models, a novel

ensemble approach is explored by combining

VGG16, ResNet, and EfficientNet. To increase

overall prediction accuracy, ensemble learning

combines several models, each of which has unique

strengths: ResNet handles deep architectures,

EfficientNet scales network dimensions effectively,

and VGG16 is excellent at extracting detailed

features. This ensemble model aims to leverage these

strengths to enhance classification performance.

Images illustrating various degrees of air pollution

are included in the dataset utilized in this study, which

is arranged according to the Air Quality Index (AQI).

More accurate forecasts were made for high pollution

levels like Hazardous/Severe by the ensemble model,

which also performed better, especially when it came

to differentiating between closely similar air quality

categories like Unhealthy and Very Unhealthy.

Complex patterns in the data can be effectively

captured by combining ensemble learning and

transfer learning.

In summary, transfer learning models, especially

when combined in an ensemble approach,

significantly improve the accuracy of air pollution

prediction systems. This method holds promise for

real-world applications, enabling more reliable air

quality monitoring and timely public health

interventions. The results of this study demonstrate

how cutting-edge deep learning methods can be

applied to urgent environmental issues like air

pollution.

2 RELATED WORKS

Samad et al. (Samad, Garuda, et al. , 2023) sought to

replace conventional air quality monitoring systems

with a new method for predicting air pollution by

utilizing machine learning-powered virtual

monitoring stations. This method utilizes data from

limited physical monitoring stations and expands its

coverage through predictive models, enabling

accurate and cost-effective air quality monitoring

over large areas.

Kumar and Pande (Kumar, Pande, et al. , 2023)

investigated the use of machine learning methods for

air pollution prediction in their case study on Indian

cities. The study underlined how crucial it is to

customize models for particular areas because

pollution sources and urban features vary widely

throughout India. Their findings highlighted the

significance of region-specific solutions in

addressing air pollution challenges.

Karimi et al. (Karimi, Asghari, et al. , 2023)

proposed white-box machine learning models for

predicting air pollution levels near industrial zones.

Unlike black-box models, their approach provided

interpretability, enabling better understanding of the

factors influencing pollution. Their work underscores

the importance of transparency in predictive models

for environmental applications.

Luo and Gong (Luo and Gong, 2023) created a

hybrid ARIMA-WOA-LSTM model for predicting

air pollutants. By combining ARIMA for temporal

pattern extraction, Whale Optimization Algorithm

(WOA) for parameter optimization, and LSTM for

deep learning-based trend prediction, this approach

demonstrated enhanced accuracy in forecasting air

quality trends.

Yang et al. (Yang, Wang, et al. , 2023) presented

a district-level air pollution forecasting system that

not only predicts pollutant levels but also evaluates

the associated health impacts and economic costs.

This comprehensive system integrates prediction

with actionable insights for policymakers,

emphasizing the multidimensional implications of air

pollution.

Pan et al. (Pan, Harrou, et al. , 2023) compared

various machine learning techniques for predicting

ozone pollution. Their work evaluated the

performance of various algorithms, highlighting the

advantages and limitations of each. This study

contributes to the selection of appropriate methods

for specific air pollution prediction tasks.

Gupta et al. (Gupta, Mohta, et al. , 2023) looked

into predicting the Air Quality Index (AQI) using a

variety of machine learning algorithms. The

usefulness of various algorithms was revealed by

their comparison study, which also emphasized the

significance of selecting appropriate models

according to the kind of air contaminants and the

properties of the data.

Li and Jiang (Li and Jiang, 2023) introduced a

novel predictive framework combining Temporal

Convolutional Networks (TCN), BiLSTM, and

DMAttention with STL decomposition. This

approach decomposed the time series into multiple

components, improving the predictive accuracy for

INCOFT 2025 - International Conference on Futuristic Technology

air pollutant concentrations by addressing seasonal

and trend variations effectively.

A study by Hardini et al. (Hardini, Chakim, et al.

, 2023) investigated the use of convolutional neural

networks (CNNs) for image-based air quality

prediction. Using visual data, their approach

classified air quality into categories such as "Good,"

"Moderate," and "Hazardous." This innovative

method leverages the potential of computer vision in

environmental monitoring.

Cao et al. (Cao, Zhang, et al. , 2023) used

Empirical Mode Decomposition (EMD) to create a

hybrid air quality prediction model. Their approach

broke down complex pollutant signals into simpler

components, allowing for more precise predictions.

This work highlights the role of signal processing

techniques in enhancing air pollution modeling.

3 PROPOSED WORK

The proposed system for air pollution prediction

utilizes transfer learning, employing pre-trained deep

learning models to categorize air quality into six

levels: Good, Moderate, Unhealthy for Sensitive

Groups, Unhealthy, Very Unhealthy, and Hazardous

or Severe. This system employs a combination of

VGG16, ResNet, and EfficientNet, which are known

for their high performance in image recognition tasks.

These models, when fine-tuned with a specific air

quality image dataset, exhibit the ability to effectively

classify the air quality levels, thus making it a

powerful tool for air pollution prediction. The dataset

used in this system contains images that represent

different levels of air pollution. These images are

categorized into six classes, each corresponding to a

specific air quality index (AQI). The dataset is

essential for deep learning model training and

validation. The dataset is preprocessed through

scaling, normalization, and augmentation to improve

efficiency and help the model generalize well to new

data. In instance, by adding variances to the training

images, data augmentation enables the model to learn

more robust features.

Transfer learning is a key aspect of the proposed

system. Rather than starting from scratch when

training a deep neural network, pre-trained models

like VGG16, ResNet, and EfficientNet are utilized.

These models can recognize intricate patterns in

visual data since they have already been trained on

extensive image datasets such as ImageNet. By using

these pre-trained networks, the system can

significantly reduce training time while maintaining

high accuracy. The models are then fine-tuned on the

air quality image dataset, allowing them to learn

domain-specific features for classifying air quality

levels accurately. This system uses VGG16 as a basis

model for feature extraction because of its ease of use

and efficacy in image categorization. Deeper

networks can train more effectively thanks to

ResNet's residual blocks, which also help with the

vanishing gradient issue. EfficientNet, a more recent

model, provides an excellent trade-off between

accuracy and computational efficiency. When

compared to individual models, the ensemble

technique, which combines various models, has

demonstrated better performance. Increased

classification accuracy results from the ensemble

model's ability to extract a greater variety of features

from the pictures.

The system follows a three-step approach as given

in the figure 1: feature extraction, fine-tuning, and

classification. Initially, relevant features are extracted

from the images by the pre-trained algorithms. A

classifier is then trained using these features and

taught to link the retrieved features to the relevant air

quality category. The pre-trained models are fine-

tuned to fit the particular task of predicting air quality.

This step is crucial as it allows the models to transfer

their learned knowledge from general image

classification tasks to air pollution-specific tasks. The

ensemble approach, which combines VGG16,

ResNet, and EfficientNet, further enhances the

Figure 1: Proposed model

Ensemble Transfer Learning for Air Quality Classiﬁcation: A Robust Model for Environmental Monitoring

prediction accuracy. By leveraging the strengths of

each model, the system can achieve higher accuracy

in classifying air quality levels. The models

complement each other in terms of capturing different

aspects of the data, such as fine-grained textures and

high-level patterns. The outputs of the three models

are averaged to get the final forecast, which

guarantees a more accurate and dependable

classification.

Additionally, the system has performance

evaluation metrics that aid in evaluating the quality of

the predictions, including accuracy, precision, recall,

and F1-score. These metrics are essential for

understanding how well the system performs in

different air quality categories, particularly for

distinguishing between more similar classes like

"Moderate" and "Unhealthy for Sensitive Groups."

Since real-time air quality monitoring is crucial in

metropolitan settings, the system can be implemented

in a variety of settings. Drones, environmental

sensors, satellite photography, and other image-based

data sources can all be integrated with the model.

Authorities and citizens can take the necessary steps

to lessen the negative consequences of poor air

quality by using the system's ability to estimate air

quality levels based on visual data to deliver timely

alerts and insights. Overall, the suggested solution

shows how effective deep learning and transfer

learning models are at accurately predicting air

pollution levels. The combination of VGG16,

ResNet, and EfficientNet in an ensemble approach

offers a state-of-the-art solution for air quality

classification, with potential applications in

environmental monitoring, urban planning, and

public health management.

4 MODULE DESCRIPTION

4.1 Data Handling:

The first module focuses on gathering and pre-

processing the air quality image dataset. Images are

collected from various sources representing air

quality at different times and locations. Good,

Moderate, Unhealthy for Sensitive Groups,

Unhealthy, Very Unhealthy, and Hazardous/Severe

are the six classifications into which these photos are

divided. To make sure the dataset is prepared for

model training, pre-processing procedures like

scaling, normalization, and data augmentation are

carried out. By increasing the dataset's variability

using data augmentation techniques like rotation,

flipping, and zooming, the model's capacity to

generalize well to new data is enhanced.

Figure 2: Air pollution Dataset’s

4.2 Model Architecture:

In the second module, the predictive model's

architecture and design are covered. VGG16, ResNet,

and EfficientNet are a trio of pre-trained deep

learning models that are used in this system. In order

to capture both low-level and high-level information

from images, each of these models has been pre-

trained on extensive datasets such as ImageNet. To fit

the task of classifying air quality, the models are

adjusted. VGG16 handles basic feature extraction,

ResNet improves the learning of deep features

through residual connections, and EfficientNet

optimizes both accuracy and computational

efficiency. The outputs of these models are then

combined in an ensemble approach to make a final

prediction.

4.3 Extraction and Fine-tuning of

Features:

Feature extraction and fine-tuning are the next steps

after defining the model architecture. In this stage,

significant features are extracted from the air quality

images by the pre-trained models. Fine-tuning is

performed on top of the pre-trained layers, allowing

the models to adapt the learned features to the air

pollution classification task. By modifying the final

layers of the models, they are tailored to predict the

six air quality categories.

INCOFT 2025 - International Conference on Futuristic Technology

4.4 Ensemble Learning:

Ensemble learning is used in the fourth module to

enhance the model's functionality. The system

leverages the unique strengths of each model by

integrating the predictions of VGG16, ResNet, and

EfficientNet. Simple patterns are the emphasis of

VGG16, deeper features are captured by ResNet,

while accuracy and efficiency are balanced by

EfficientNet. By averaging the results of these

models, the final prediction is produced, lowering the

possibility of inaccuracies resulting from the

shortcomings of any one model. When predicting air

quality levels, an ensemble technique guarantees

greater accuracy and robustness.

4.5 Model Evaluation:

The last module entails assessing the ensemble

model's performance and getting it ready for

deployment. Evaluation metrics are computed to

evaluate the model's performance across the six air

quality categories, including accuracy, precision,

recall, and F1-score. These measurements offer

information about the system's advantages and

shortcomings. When the model performs well

enough, it is used to predict air quality in real time. In

metropolitan settings, the system can be used to offer

timely air quality alerts and predictions by integrating

with environmental monitoring tools like satellite

images or sensors. This helps with public health and

safety decision-making.

5 IMPLEMENTATION

The implementation of the air pollution prediction

system involves several steps, including data

preparation, model training, fine-tuning, and

evaluation. The primary focus is on the use of transfer

learning (TL) models—VGG16, ResNet, and

EfficientNet—each of which has specific strengths in

image recognition tasks. These models are well-

equipped to tackle novel, domain-specific tasks like

air pollution prediction since they have already been

pre-trained on big datasets like ImageNet. The

following sections describe each model in detail,

outlining its role in the overall system

implementation.

5.1 VGG16 (Visual Geometry

Group16)

VGG16 is a deep convolutional neural network

(CNN) with 16 weight layers that was created by

Oxford's Visual Geometry Group. Its design, which

is built on tiny (3x3) convolutional filters, pooling

layers, and fully linked layers, is well known for

being straightforward and efficient. VGG16 is

appropriate for a variety of picture classification

problems because it is excellent at capturing low- and

mid-level visual features like edges, textures, and

patterns. In this system, VGG16 shown in Figure 3 is

used as the base model for feature extraction. After

pre-training on ImageNet, the model is fine-tuned

using air quality images, enabling it to recognize

specific features associated with different levels of

pollution. The simplicity and ease of implementation

make VGG16 a good starting point in this air quality

classification system.

Figure 3: VGG16 Model Architecture

5.2 ResNet (Residual Networks)

ResNet, a product of Microsoft Research, uses

residual learning to solve the problem of training

extremely deep neural networks. The disappearing

gradient issue is lessened by using skip connections,

also known as shortcuts, which make gradients flow

more readily during backpropagation. ResNet models

can be built with various depths, such as ResNet-50,

ResNet-101, and ResNet-152 shown in Figure 4. In

this system, ResNet is used to capture deeper and

more complex features.

Figure 4: ResNet50 Model Architecture

Ensemble Transfer Learning for Air Quality Classiﬁcation: A Robust Model for Environmental Monitoring

5.3 EfficientNet

EfficientNet, a recent architecture proposed by

Google AI, optimizes both accuracy and

computational efficiency. It uses a complex scaling

technique to balance the model's depth, width, and

resolution. In contrast to previous models like

VGG16 and ResNet, EfficientNet is renowned for

obtaining great accuracy with fewer parameters. In

the proposed system, EfficientNet shown in Figure 5

serves as the model for efficient image classification,

offering excellent performance without requiring

excessive computational resources.

Figure 5: EfficientNet Model Architecture

5.4 Ensemble Model (Combination of

VGG16, ResNet, and EfficientNet)

The proposed system combines VGG16, ResNet, and

EfficientNet in an ensemble learning approach. Each

model has its strengths, and by combining their

outputs, the system leverages the strengths of each

architecture to enhance overall performance. VGG16

captures basic features, ResNet learns deeper, more

complex patterns, and EfficientNet optimizes both

accuracy and computational efficiency. In this

ensemble setup, each model processes the input

image independently and makes a prediction. By

averaging the results from all three models, the final

prediction is produced, making the system less

vulnerable to mistakes that could result from the

shortcomings of any one model. This approach results

in a more robust and accurate air quality classification

system.

6 RESULTS AND DISCUSSION

The evaluation of the air pollution prediction system,

which makes use of transfer learning using VGG16

and ResNet models, is presented in the results and

discussion section. An air quality picture dataset that

was divided into six different classes—Good,

Moderate, Unhealthy for Sensitive Groups,

Unhealthy, Very Unhealthy, and

Hazardous/Severe—was used to train and assess the

system. To evaluate how well the suggested model

worked, the main performance metrics—accuracy,

precision, recall, and F1-score—were calculated.

Furthermore, a comparison between the ensemble

technique and individual models (ResNet and

VGG16) is provided.

6.1 Performance Metrics

Using a variety of performance criteria, the ensemble

model—which blends VGG16 and ResNet—was

assessed on a test dataset. Compared to the individual

models, the ensemble model's accuracy was

noticeably higher. As illustrated in Figure 6, the

ensemble model obtained an accuracy of 92.1%,

whereas VGG16 and ResNet obtained 73.23% and

84.62%, respectively. These outcomes unequivocally

show the benefit of the ensemble technique, which

considerably increases classification accuracy over

ResNet alone by successfully capturing both low-

level and high-level information from the images.

Even though VGG16 performed well, the ensemble

strategy outperformed it, highlighting the advantages

of integrating these two powerful models.

For every class, F1-score, precision, and recall

metrics were also computed. With precision, recall,

and F1-scores averaging around 0.91, 0.92, and 0.91,

respectively, the ensemble model demonstrated a

well-balanced performance across all classes. By

reducing false positives and false negatives, this

shows that the algorithm can accurately predict the air

quality categories. Since precise classification is

critical for public health and safety in real-world

applications, the model's ability to balance precision

and recall is shown by the higher F1-score.

Figure 6: Comparison with other models

100

VGG16 ResNet Ensemble

Model

Accuracy

INCOFT 2025 - International Conference on Futuristic Technology

6.2 Class-wise Performance Analysis

A deeper analysis of the results revealed that the

system performed exceptionally well in predicting

"Good" and "Moderate" air quality levels, achieving

accuracy rates of 94% and 91%, respectively.

However, the system showed slightly lower

performance in predicting "Hazardous/Severe"

levels, with an accuracy of 85%. This may be due to

the limited number of images in the

"Hazardous/Severe" category, making it more

challenging for the model to accurately learn and

classify these extreme cases. Despite this, the

ensemble model maintained an overall high level of

performance across all air quality categories,

demonstrating its robustness and versatility.

6.3 Comparison with Other

Approaches

The ensemble model’s performance was compared to

other studies in the domain of air pollution prediction.

In comparison with (Samad et al., 2023), where

machine learning algorithms were employed to

predict air pollution using sensor data, the proposed

system demonstrated superior accuracy by leveraging

image-based classification with deep learning

techniques. In (Kumar and Pande, 2023), machine

learning approaches were used with limited feature

extraction methods, while the ensemble model

benefits from advanced convolutional architectures

that can capture complex and meaningful features

from images, thereby achieving higher classification

performance. In contrast to (Luo and Gong, 2023),

which relied on ARIMA-WOA-LSTM models for

time-series prediction, the current image-based

approach provides real-time predictions, making it

suitable for integration into smart city infrastructure.

The ensemble model has demonstrated a more

efficient and scalable solution for air quality

prediction based on visual data.

6.4 Limitations and Future Work

Despite the promising results, several limitations

persist. One key limitation is the reliance on image-

based data, which might not always be available,

particularly in areas with fewer monitoring stations.

Additionally, the model’s performance on the

"Hazardous/Severe" category could be further

improved by acquiring more images in this category

to overcome the challenge of class imbalance. A

potential improvement would be to implement data

augmentation techniques to artificially expand the

dataset for rare classes, which may help in improving

the model's predictive capabilities for extreme air

pollution levels. Future work could also focus on

integrating additional transfer learning models such

as DenseNet, which is known for high performance

in image classification tasks, to further enhance the

prediction accuracy. Combining the ensemble

approach with sensor-based data or time-series data

could also increase the robustness of the system by

allowing it to handle both visual and temporal data,

providing more comprehensive and accurate

predictions.

7 CONCLUSIONS

In order to classify air quality levels into six different

categories—Good, Moderate, Unhealthy for

Sensitive Groups, Unhealthy, Very Unhealthy, and

Hazardous/Severe—this paper presents an air

pollution prediction system that makes use of transfer

learning techniques, specifically using VGG16 and

ResNet models. The suggested method shows how

deep learning models can be used to reliably forecast

air pollution levels from visual input. The system

outperforms individual models by using the

capabilities of both VGG16 and ResNet, achieving an

impressive accuracy of 92.1%.

According to the experimental findings, the

ensemble model exhibits remarkable performance in

every air quality class, with a well-balanced

precision, recall, and F1-score that guarantees

excellent prediction reliability. Even when faced with

obstacles like class imbalance, the model's

performance held up well, but it could still be

improved, especially when it came to predicting high

pollution levels like "Hazardous/Severe." Future

research will concentrate on growing the dataset,

adding more transfer learning models, and using

time-series or sensor-based data to improve

prediction accuracy even more. The suggested system

has important applications in the real world. It

provides a scalable and effective way to monitor air

quality in real time and can help with environmental

and public health policy decision-making. This

technology can improve the general quality of life in

urban settings by reducing the hazards related to poor

air quality by offering precise and timely predictions.

In summary, the application of transfer learning to

the prediction of air pollution not only advances the

expanding field of environmental monitoring but also

demonstrates the revolutionary potential of deep

learning technologies in addressing global issues such

as climate change and air pollution.

Ensemble Transfer Learning for Air Quality Classiﬁcation: A Robust Model for Environmental Monitoring

REFERENCES

Samad, A., Garuda, S., Vogt, U., & Yang, B. (2023). Air

pollution prediction using machine learning

techniques–an approach to replace existing monitoring

stations with virtual monitoring stations. Atmospheric

Environment, 310, 119987.

Kumar, K., & Pande, B. P. (2023). Air pollution prediction

with machine learning: a case study of Indian cities.

International Journal of Environmental Science and

Technology, 20(5), 5333-5348.

Karimi, S., Asghari, M., Rabie, R., & Niri, M. E. (2023).

Machine learning-based white-box prediction and

correlation analysis of air pollutants in proximity to

industrial zones. Process Safety and Environmental

Protection, 178, 1009-1025.

Luo, J., & Gong, Y. (2023). Air pollutant prediction based

on ARIMA-WOA-LSTM model. Atmospheric

Pollution Research, 14(6), 101761.

Yang, W., Wang, J., Zhang, K., & Hao, Y. (2023). A novel

air pollution forecasting, health effects, and economic

cost assessment system for environmental

management: From a new perspective of the district-

level. Journal of Cleaner Production, 417, 138027.

Pan, Q., Harrou, F., & Sun, Y. (2023). A comparison of

machine learning methods for ozone pollution

prediction. Journal of Big Data, 10(1), 63.

Gupta, N. S., Mohta, Y., Heda, K., Armaan, R., Valarmathi,

B., & Arulkumaran, G. (2023). Prediction of air quality

index using machine learning techniques: a

comparative analysis. Journal of Environmental and

Public Health, 2023(1), 4916267.

Li, W., & Jiang, X. (2023). Prediction of air pollutant

concentrations based on TCN-BiLSTM-DMAttention

with STL decomposition. Scientific Reports, 13(1),

4665.

Hardini, M., Chakim, M. H. R., Magdalena, L., Kenta, H.,

Rafika, A. S., & Julianingsih, D. (2023). Image-based

air quality prediction using convolutional neural

networks and machine learning. Aptisi Transactions on

Technopreneurship (ATT), 5(1Sp), 109-123.

Cao, Y., Zhang, D., Ding, S., Zhong, W., & Yan, C. (2023).

A hybrid air quality prediction model based on

empirical mode decomposition. Tsinghua Science and

Technology, 29(1), 99-111.

INCOFT 2025 - International Conference on Futuristic Technology