Predecting Adverse Drug Reactions with XGBoost a

Pharmacovigilance Application

K. Jnana Sadhana, Veera Raghavan J., Bellamgubba Anoch, P. Kiran Sree,

Raja Rao P. B. V. and B. Satyanarayana Murthy

Department of CSE, Shri Vishnu Engineering College for Women, Vishnupur, Bhimavaram, Andhra Pradesh, India

Keywords: Adverse Drug Reactions (ADRs), Machine Learning(ML) in Pharmacovigilance, XGBoost for ADR

Detection, Feature Selection, Explainability(SHAP), AI in Drug Safety, Healthcare.

Abstract: Adverse Drug Reactions (ADRs) pose a significant challenge to modern pharmacovigilance, leading to severe

health implications, increased healthcare costs, and regulatory concerns. Traditional ADR detection methods

rely heavily on manual pharmacovigilance and rule-based expert systems, which are slow, subjective, and

limited in scalability.However, existing ML models either suffer from overfitting, lack of generalizability, or

black-box limitations.This study proposes a novel XGBoost-based ADR prediction model that achieves 91%

accuracy, outperforming traditional classifiers (Naïve Bayes, SVM, Random Forest) and deep learning

models (Graph Neural Networks, Neural Collaborative Filtering, Deep Ensembles) found in literature. The

proposed model uses feature engineering with SHAP explainability, class balancing techniques, and

hyperparameter tuning to improve predictive performance. By leveraging Therapeutic Class, Action Class,

and Chemical Properties, the model not only enhances accuracy but also ensures interpretability, making it

suitable for real-world clinical decision systems.Experimental results demonstrate that the optimized

XGBoost model achieves 91% accuracy, 92% precision, and 91% recall, making it a competitive alternative

to deep learning-based ADR detection models.

1 INTRODUCTION

1.1 Background & Motivation

ADRs are an important source of concern in modern

medicine. impacting the lives of millions of patients

worldwide. Such adverse and damaging

consequences of pharmaceutical medications are not

only detrimental to patient safety, but also lead to

higher hospitalization rates, healthcare costs, and

regulatory hurdles. Due to the integration of more

similar therapeutic agents, the complexity of drug

interactions has dramatically increased. The advent of

personalised medicine and polypharmacy the use of

multiple drugs concurrently has made it even more

imperative to have a strong and automated system

that can help prevent potential ADR before they

become clinically apparent (N. Ibrar, I. Hamid, and

Q. Nawaz., 2023). Thus, the computational prediction

and prevention of ADRci has evolved into a vital

segment of pharmacovigilance.

1.2 Artificial Intelligence's Function in

ADR Detection

The advancements in machine learning (ML) and

artificial intelligence (AI) have altered many fields,

and pharmacovigilance is no different. While

traditional approaches to ADR detection depend

primarily on manual reporting systems, clinical trials,

and rule-based expert reviews, recent advancements

see AI methodologies capable of analyzing large data

sets, identifying non-obvious patterns, and predicting

adverse reactions with greater accuracy Other

machine learning algorithms such as Support Vector

Machines (SVMs), Random Forest Pokkuluri, K.S. et

al. (2025), and Naive Bayes classifiers are found to

be more effective in detecting ADRs due to the

patient history as well as the chemical properties of

the drugs.

1.3 Significance of This Research

This study seeks to optimize the XGBoost-based

ADR (Saravanan, D., Arunkumar, G., Ragupathi,

Sadhana, K. J., J., V. R., Anoch, B., Sree, P. K., P. B. V., R. R. and Murthy, B. S.

Predecting Adverse Drug Reactions with XGBoost a Pharmacovigilance Application.

DOI: 10.5220/0013871700004919

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

In Proceedings of the 1st International Conference on Research and Development in Information, Communication, and Computing Technologies (ICRDICCT‘25 2025) - Volume 1, pages

713-720

ISBN: 978-989-758-777-1

713

T. et al., 2025) prediction model by proposing the

following objectives:

Use feature selection techniques to improve

prediction accuracy. Enhancement of model

interpretability through SHAP-based Feature

Importance Analysis Merge Therapeutic Class,

Chemical Class, and Action Class to enhance feature

representation. Make scalable for practical

application in pharmacovigilance systems. This work

aims to derive a black-box-free, high-performing

model and subsequently help deliver a strong AI-

enabled pharmacovigilance structure that can assist

doctors recognise potential drug adverseevents and

avert fatal drug combinations ahead of time. This

study's results can play an important role in the future

of AI-based medication safety monitoring that

enables data-driven healthcare decision-making and

precision medicine.

2 RELATED WORK

Adverse drug reaction (ADR) is one of the most

problematic complications in modern pharmacology

that challenge the drug deduction process and patient

safety. Conventional methods of identifying adverse

drug reactions (ADRs), including clinical trials and

post-market monitoring, are rarely successful in

detecting infrequent or late-emerging adverse effects

prior to a drug reaching the market. The introduction

of machine learning (ML) and artificial intelligence

(AI) has resulted in the emergence of predictive

models that can be used to detect ADRs earlier in the

drug development process. We next review existing

ML-based ADR prediction methods, their successes,

and their shortcomings.

2.1 Using Machine Learning to Predict

ADRs

Several machine learning techniques have been

applied to pre-market ADR prediction, including

graph-based models, matrix factorization, neural

collaborative filtering, and deep learning approaches.

2.1.1 Graph Neural Networks (GNNs) for

Drug-Drug Interaction ADR

Prediction

Though graph-based learning methods have been

utilized for advancing ADR prediction by modelling

drug-drug interactions (DDIs). Chandra

Umakantham et al. proposed a GNN-based model

that performed self-supervised learning to identify

ADRs from DDIs. 2024. greatly enhancing

predictive accuracy compared to traditional statistical

methods. Inspired by this observation, Patel & Patel

(2024) took a step further, incorporating causal

inference methods with the GNNs to make ADR

predictions more robust and interpretable.

Limitations: While GNNs offer improved

accuracy, they require large-scale graph

representations and high computational resources,

making them less feasible for real-time applications.

2.1.2 Signed Networks and Matrix

Factorization for ADR Prediction

Alternative approaches leverage network structures

and probabilistic dependencies to enhance ADR

detection.A signed network-based method was

presented by Zhuang & Wang (2021), who used the

topological structures of pharmacological networks

to infer possible adverse drug reactions.A

probabilistic matrix tri-factorization model was

presented by Zhu et al. in 2021,which improves

interpretability by considering ADR dependencies

across drug pairs.

Limitations: These methods depend on data

completeness, making them less effective in sparse

datasets where missing interactions occur.

2.1.3 Neural Collaborative Filtering and

Hybrid ML Approaches

To address the cold-start problem in ADR prediction

(i.e., handling novel drugs with little existing data),

recent studies have applied collaborative filtering

techniques.Xiong et al. (2023) designed a Neural

Collaborative Filtering (NCF) model, leveraging drug

feature similarities to improve ADR prediction in

unseen drugs.Ibrar et al. (2023) implemented a hybrid

ML model, combining deep learning and structured

feature engineering for ADR classification.

Limitations: Despite their generalization ability,

collaborative filtering models struggle with rare ADR

cases as they rely heavily on historical data.

2.1.4 Semi-Supervised Learning for ADR

Prediction

To improve generalization across different drug

datasets, researchers have adopted semi-supervised

learning techniques.Yan et al. (2022) developed a

similarity network-based semi-supervised learning

model, enhancing generalizability across multiple

ADR datasets.

Limitations: Semi-supervised learning methods

require high-quality unlabeled data, and their

ICRDICCT‘25 2025 - INTERNATIONAL CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION,

COMMUNICATION, AND COMPUTING TECHNOLOGIES

714

performance depends on data augmentation

techniques.

2.2 Structure-based ADR Prediction

Models

2.2.1 Relationships between Structure and

Activity (SAR) and Quantitative

Structure and Activity (QSAR)

ModelsStructure-based models analyze the chemical

composition of drugs to infer their likelihood of

causing ADRs.Kang et al. (2022) applied Support

Vector Machines (SVMs) in an SAR model to

forecast antiseizure drug teratogenicrisk.Zhou et al.

(2021) used Random Forest to create a QSAR model

that evaluated the degree of drug-induced

rhabdomyolysis(PBV et al., 2024).

Limitations: SAR and QSAR models require

extensive feature engineering, and their accuracy is

highly dependent on curated chemical datasets.

2.2.2 Structural Alerts for ADR Risk

Identification

Structural alert models identify toxicity risk markers

in drug molecules.Long et al. (2023) implemented a

structural alert-based model for predicting drug-

induced QT prolongation, improving the architecture

of the Adverse Outcome Pathway (AOP).

Limitations: These models require manual rule

formulation, making them less adaptable to novel

drug compounds.

2.3 Deep Learning for ADR Prediction

2.3.1 NLP-Driven ADR Susceptibility

Prediction via Chemical Language

Models

Recent studies have applied Natural Language

Processing (NLP) models to Extractive text-based

drug safety reports. To enhance ADR detection

through preclinical drug screening, Lin et al. (2024)

proposed a deep chemical language model for the

review of drug safety.

Limitations: NLP-based models are data biases

prone which need large scale corpus, textual data for

training.

2.3.2 Risk Prediction of Drug Induced Liver

Injury (DILI) and Cardiotoxicity

Deep learning algorithms have also been employed to

predict ADRs including cardiotoxicity and drug-

induced liver injury (DILI)Minerali et al. (2020)

reported that deep learning performed better than the

traditional ML methods for predicting the DILI cases

Pokkuluri, K.S. et al. (2025), (Sree et al., 2024).

2.4 Comparative Analysis of ADR

Prediction Models

Table 1 Shows the summarizes the performance

metrics for various models applied to ADR

prediction.

2.5 Challenges in Traditional ADR

Detection

Despite advancements in pharmacovigilance,

traditional ADR detection methods face several

challenges:

2.5.1 Underreporting and Data Bias

Over 90% of ADR cases remain unreported due to

lack of awareness, reporting delays, and legal

concerns.ADR reports are disproportionately skewed

towards widely prescribed drugs, making rare ADRs

harder to detect.

2.5.2 Scalability Issues in Rule-Based

Methods

Traditional rule-based expert systems require

frequent manual updates to accommodate newly

introduced drugs.These systems fail to adapt to

complex drug-drug interactions (DDIs) and emerging

adverse reactions.

2.5.3 Limitations of Classical Machine

Learning Models

Conventional models like SVMs, Decision Trees, and

Naïve Bayes struggle with high-dimensional ADR

datasets.Class imbalance in ADR datasets results in

biased predictions, where models favor commonly

occurring reactions while ignoring rare but critical

ADRs.

Predecting Adverse Drug Reactions with XGBoost a Pharmacovigilance Application

715

Table 1: Summarizes the Performance Metrics for Various Models Applied to ADR Prediction.

Approach Key Contribution Dataset Used Accuracy Limitations

GNN-Based Model

(ChandraUmakantha

m et al., 2024)

Improved ADR

detection in DDIs

FAERS, SIDER 89%

Requires high

computational resources

Causal Inference +

GNNs (Patel & Patel,

2024

)

Enhanced reliability

with causal reasoning

FAERS 85% Limited interpretability

Probabilistic Matrix

Factorization (Zhu et

al., 2021)

Improved ADR

interpretability

FAERS 83%

Requires high-quality

labeled data

Neural Collaborative

Filtering (Xiong et al.,

2023

)

Generalized ADR

prediction for new

dru

FAERS, Open TG-

GATEs

87% Struggles with rare ADRs

SAR Model (Kang et

al., 2022)

Predicted teratogenic

risk

Open TG-GATEs 86%

High feature engineering

complexity

Structural Alert

Model (Long et al.,

2023)

Identified

cardiotoxicity risk

DrugBank 84%

Requires manual rule

formulation

Deep Learning

Ensemble (Karim et

al., 2021)

Enhanced

cardiotoxicity

prediction

DrugBank 90%

Black-box nature limits

explainability

Ensembles achieve high accuracy (~90%), their

"black-box" nature limits their use in clinical

settings.Clinicians require transparent and

explainable AI models for real-world adoption, which

deep learning models often fail to provide.

3 PROPOSED METHODOLOGY

Dataset is downloaded from kaggle repository.named

with medicine datset.CSV

Step 1: Data Collection & Preprocessing

• Dataset: Extracted from structured medical

records, clinical trial data, and publicly

available ADR datasets.

• Data Cleaning: Removal of duplicates,

handling missing values, and filtering out

irrelevant features.

• Feature Selection:

• Important attributes: Chemical Class,

Therapeutic Class, Side Effects, Substitutes,

and Action Class.

• Unimportant attributes (e.g., common

stopwords, noise) were eliminated to

improve model efficiency.

Step 2: Feature Engineering & Representation

• Text Features: Side effects and drug

reactions were processed using TF-IDF and

NLP-based techniques.

• Categorical Features: Encoded using Label

Encoding & One-Hot Encoding.

• Numerical Features: Standardized using

Min-Max Scaling.

• Feature Selection: Used SHAP Analysis and

XGBoost Feature Importance to retain high-

impact features.

Step 3: Model Selection & Training

• To identify the best-performing model,

various machine learning classifiers were

experimented with:

• Baseline Models: Naive Bayes: (Accuracy

~37%) – Struggled with imbalanced data.

• SVM (Support Vector Machine):

Computationally expensive, slow training.

• Random Forest: Achieved 99% accuracy,

but overfitted to training data.

• Advanced Models: XGBoost: Achieved

91% accuracy after hyperparameter tuning.

• Stacking Ensemble: (Combination of

Random Forest & XGBoost) boosted

performance.

ICRDICCT‘25 2025 - INTERNATIONAL CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION,

COMMUNICATION, AND COMPUTING TECHNOLOGIES

716

• Deep Learning (Surveyed in Literature

Review):

• Graph Neural Networks (GNNs): High

accuracy (89%) but required extensive

computational resources.

• Neural Collaborative Filtering (NCF): 87%

accuracy but struggled with rare ADR

detection.

• Deep Ensemble Networks: 90% accuracy,

but interpretability was a major concern.

Step 4: Hyperparameter Optimization

• To enhance performance, hyperparameters

were tuned using:Grid Search CV – To find

the best combination of learning rate, depth,

and regularization.

• Regularization (lambda, alpha) – To prevent

overfitting.

• Balanced Sampling Strategy – Adjusting

scale_pos_weight for rare ADR detection.

Step 5: Model Evaluation & Explainability

Evaluation Metrics Used:

• Accuracy = 91%

• Precision = 92%

• Recall = 91%

• F1-score = 91%

Explainability with SHAP

Identified the most influential features (Therapeutic

Class, Action Class, Side Effects).

Ensured transparency by visualizing feature

importance in individual predictions.

Confusion Matrix Analysis:

Reduced misclassification of rare ADR classes.

Improved recall for underrepresented classes like

Class 3, 7, and 16.

Step 6: Deployment & Future Enhancements

• Model Deployment: The final XGBoost

model was saved as xgboost_91.pkl, ready

for integration into clinical decision

systems.

• Ensemble Learning: Potential stacking with

GNNs + XGBoost for even better

performance.

• Continuous Learning: Incorporate new

patient data into incremental retraining

pipelines, enabling the model to adapt to

evolving disease patterns and demographics.

• Security & Privacy: Ensure HIPAA/GDPR

compliance with strong encryption,

anonymization, and secure API endpoints

during deployment.

Figure 1: Workflow Diagram ADR Prediction Model.

We introduce a pipeline starting from data collection

and data preprocessing, SHAP based feature

selection, followed by class balancing. Once

deployed, the XGBoost model is trained, optimized,

and evaluated. This figure 1 displays the overall

workflow of ADR prediction.

The pipeline further incorporates hyperparameter

tuning for performance maximization, cross-

validation to ensure generalizability, and model

interpretability tools to aid clinical decision-making.

As shown in Table 2, the proposed model

outperforms the baseline methods in ADR prediction.

Figure: 2 System Architecture for AI-Driven ADR

Prediction.

Predecting Adverse Drug Reactions with XGBoost a Pharmacovigilance Application

717

The Figure 2 system architecture outlines the core

components of an AI-based ADR detection

framework. It starts with data acquisition from drug

and patient records, followed by feature engineering,

machine learning-based classification, evaluation

using performance metrics, and deployment into

clinical decision support systems.

4 RESULTS AND DISCUSSION

The performance evaluation of various machine

learning models used for adverse drug reactions

(ADRs) is presented in this section.Prediction and

discusses the impact of the proposed XGBoost-based

optimized model.

4.1 Model Performance Comparison

A detailed comparison of the evaluated models

including Logistic Regression, Random Forest, SVM,

and Gradient Boosting methods is conducted. The

XGBoost-based optimized model consistently

outperforms the others, achieving the highest ROC-

AUC and F1-score, indicating a strong balance

between sensitivity and specificity.

Table 2: Performance Comparison of ADR Prediction Models.

Model Accurac

Precision Recall F1-Score Remarks

Naive Bayes 37% 39.90% 37.50% 34.60%

Weak performance, fails

on imbalanced data

Random Forest 99% 99.18% 99.16% 99.16% Overfits on training data

XGBoost 91% 92% 91% 91%

Best trade-off between

accuracy & explainability

This figure 3 compares the accuracy, precision, recall,

and F1-score of many machine learning models, such

as XGBoost, Random Forest, and Naïve Bayes.

Superior accuracy and balanced recall are attained by

the optimized XGBoost model, guaranteeing

dependable ADR prediction with enhanced feature

interpretability.

Figure 2: Confusion Matrix for Xgboost-Based ADR

Prediction.

The confusion matrix presents the classification

performance of the optimized XGBoost model across

various therapeutic classes. The diagonal values

indicate correctly predicted ADR cases, while off-

diagonal values represent misclassifications. The

model demonstrates high classification accuracy,

with minimal misclassification rates, ensuring robust

drug safety assessment in pharmacovigilance.

4.2 Discussion on Model Performance

The proposed XGBoost-based optimized ADR

prediction model achieves an accuracy of 91%, which

is higher than several baseline models and

comparable to deep learning models reviewed in the

literature.

4.2.1 Performance Gains from Data

Balancing & Feature Selection

Before balancing: XGBoost showed 100% accuracy

(overfitted model), indicating bias toward majority

classes.

After balancing: Accuracy stabilized at 81%.

Feature selection (SHAP-based pruning) helped

remove noisy features, boosting accuracy to 91%.

4.2.2 Impact of Hyperparameter Tuning

Grid Search Optimization improved model

generalization.

Increased min_child_weight and gamma controlled

overfitting.Adjusting scale_pos_weight helped

increase recall for rare ADR cases.

ICRDICCT‘25 2025 - INTERNATIONAL CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION,

COMMUNICATION, AND COMPUTING TECHNOLOGIES

718

4.2.3 Comparison with Deep Learning

Models

Literature models like Graph Neural Networks

(GNNs), Neural Collaborative Filtering (NCF), and

Deep Learning Ensembles achieve ~90% accuracy.

4.3 Feature Importance Analysis

Feature selection using SHAP (SHapley Additive

Explanations) was performed to rank the most

influential attributes affecting ADR prediction.

Figure 4 shows the SHAP Interaction Analysis for

ADR Prediction.

Figure 3: Shap Interaction Analysis for ADR Prediction.

5 CONCLUSIONS

The prediction of Adverse Drug Reactions (ADRs) is

a crucial aspect of pharmacovigilance and drug

safety. Traditional methods, including spontaneous

reporting systems (SRS), clinical trials, and expert-

driven rule-based systems, have limitations in terms

of scalability, delayed detection, and subjectivity. By

facilitating real-time analysis of massive

pharmaceutical datasets, the combination of artificial

intelligence and machine learning has revolutionized

ADR detection.To effectively forecast ADRs, this

study investigated a number of machine learning

models, such as XGBoost, Random Forest, and Naïve

Bayes.The findings indicate that:Naïve Bayes

underperformed, achieving only 37.5% accuracy,

making it unsuitable for high-stakes applications in

ADR detection.Random Forest performed well, with

99.16% accuracy, but lacked interpretability.XGBo

ost (Tuned & Improved) delivered the best perform

ance, achieving a final accuracy of 91% after

addressing overfitting concerns and refining the

feature selection process.By employing SHAP-based

feature selection, the study identified highly relevant

features such as Therapeutic Class, Chemical Class,

and specific drug-induced side effects (e.g., diarrhea,

skin reactions, and hypoglycemia). This enhanced

feature interpretability, making the model suitable for

clinical decision support systems.

6 FUTURE WORK

While this study demonstrates high predictive

accuracy and interpretability, future research should

explore: Deep learning-based hybrid models,

integrating Natural language processing (NLP) and

graph neural networks (GNNs) are used to improve

ADR prediction. implementation of the suggested

approach in clinical settings to assess its effects on

medication safety management.its impact on drug

safety. Expansion of dataset sources, incorporating

electronic health records (EHRs) and patient-reported

ADRs for comprehensive pharmacovigilance

REFERENCES

A. Mohsen, L. P. Tripathi, and K. Mizuguchi, "Deep

Learning Prediction of Adverse Drug Reactions in Drug

Discovery Using Open TG-GATEs and FAERS

Databases," Frontiers in Drug Discovery, vol. 1, Oct.

2021, doi: 10.3389/fddsv.2021.768792.

A. Karim, M. Lee, T. Balle, and A. Sattar, "CardioTox Net:

A Robust Predictor for hERG Channel Blockade Based

on Deep Learning Meta-Feature Ensembles," Journal of

Cheminformatics, vol. 13, p. 60, 2021, doi: 10.1186/s1

3321-021-00541-z.

C. Yan, G. Duan, Y. Zhang, F.-X. Wu, Y. Pan, and J. Wang,

"Predicting Drug- Drug Interactions Based on Integrat

ed Similarity and Semi- Supervised Learning," IEEE/

ACM Transactions on Computational Biology and

Bioinformatics, vol. 19, no. 1, pp. 168–179, 2022, doi:

10.1109/TCBB.2020.2988018.

E. Minerali, D. H. Foil, K. M. Zorn, T. R. Lane, and S.

Ekins, "Comparing Machine Learning Algorithms for

Predicting Drug- Induced Liver Injury (DILI)," Molec

ular Pharmaceutics, vol. 17, pp. 2628–2637, 2020, doi:

10.1021/acs.molpharmaceut.0c00326.

F. Mostafa and M. Chen, "Computational Models for

Predicting Liver Toxicity in the Deep Learning Era,"

Frontiers in Toxicology, vol. 5, pp. 1340860, 2024, doi:

10.3389/ftox.2023.1340860.

J. Zhuang and H. Wang, "Drug-Drug Adverse Reactions

Prediction Based On Signed Network," 2021 11th Inte

rnational Conference on Information Technology in

Medicine and Education (ITME), 2021, pp. 277–281,

doi: 10.1109/ITME53901.2021.00064.

Predecting Adverse Drug Reactions with XGBoost a Pharmacovigilance Application

719

J. Zhu, Y. Liu, Y. Zhang, and D. Li, "Attribute Supervised

Probabilistic Dependent Matrix Tri- Factorization Mo

del for the Prediction of Adverse Drug- Drug Interacti

on," IEEE Journal of Biomedical and Health Informat

ics, vol. 25, no. 7, pp. 2820– 2832, 2021, doi: 10.110

9/JBHI.2020.3048059.

J. Lin, Y. He, C. Ru, W. Long, M. Li, and Z. Wen,

"Advancing Adverse Drug Reaction Prediction with

Deep Chemical Language Model for Drug Safety Eval

uation," International Journal of Molecular Sciences,

vol. 25, no. 8, 2024, doi: 10.3390/ijms25084516.

J. Patel and R. Patel, "Adverse Drug Reaction Prediction:

Graph Neural Networks and Causal Inference Techniq

ues," 2024 4th Interdisciplinary Conference on Electri

cs and Computer (INTCEC), pp. 1– 5, doi: 10.1109/I

NTCEC61833.2024.10603081.

Josphineleela, R., Raja Rao, P.B.V., shaikh, A. et al. A

Multi-Stage Faster RCNN-Based iSPLInception for

Skin Disease Classification Using Novel Optimization

. J Digit Imaging 36, 2210– 2226 (2023). https://doi.or

g/10.1007/s10278-023-00848-3

L. Kang, Y. Duan, C. Chen, S. Li, M. Li, L. Chen, and Z.

Wen, "Structure-Activity Relationship (SAR) Model

for Predicting Teratogenic Risk of Antiseizure

Medications in Pregnancy by Using Support Vector

Machine," Frontiers in Pharmacology, vol. 13, pp.

747935, 2022, doi: 10.3389/fphar.2022.747935.

N. Ibrar, I. Hamid, and Q. Nawaz, "Prediction of Drug

Reactions through Hybrid Techniques," 2023 9th

International Conference on Optimization and Applica

tions (ICOA), pp. 1– 6, doi: 10.1109/ICOA58279.202

3.10308824.

O. ChandraUmakantham, S. Srinivasan, and V. Pathak,

"Detecting Side Effects of Adverse Drug Reactions

Through Drug-Drug Interactions Using Graph Neural

Networks and Self- Supervised Learning," IEEE Acce

ss, vol. 12,pp.93823–93840,2024,doi:

10.1109/ACCESS.2024.3407877.

PBV, R.R., Pokkuluri, K.S., Prasad, M., Sharma, N.,

Murthy, B.N., Karunasri, A. (2025). Ensemble Fusion

for Enhanced Malicious URL Detection by Integrating

Machine Learning and Deep Learning Techniques.IC4

S 2024. Lecture Notes of the Institute for Computer

Sciences, Social Informatics and Telecommunications

Engineering, vol 597. Springer, Cham. https://doi.org/

10.1007/978-3-031-77075-3_27

Pokkuluri, K.S. et al. (2025). Generative Adversarial

Networks (GANs) for Drug Discovery.IC4S 2024.

Lecture Notes of the Institute for Computer Sciences,

Social Informatics and Telecommunications Engineeri

ng, vol 599. Springer, Cham. https://doi.org/10.1007/9

78-3-031-77081-4_21

S. Li, Z. Xu, M. Guo, M. Li, and Z. Wen, "Drug-Induced

QT Prolongation Atlas (DIQTA) for Enhancing Cardi

otoxicity Management," Drug Discovery Today, vol.

27, pp. 831– 837, 2022, doi: 10.1016/j.drudis.2021.10

.009.

Saravanan, D., Arunkumar, G., Ragupathi, T. et al. Weigh

ted Majority Voting Ensemble for MRI-Based Brain

Tumor Classification Using Capsule Networks and

XGBoost. Appl Magn Reson 56, 395–425 (2025).

https://doi.org/10.1007/s00723-024-01731-2

Sree, P.K., Usha Devi N, S., Chintalapati, P.V., Babu, G.R.,

Raja Rao, P. (2024). Drug Recommendations Using a

Reviews and Sentiment Analysis by RNN.IC4S 2023.

Lecture Notes of the Institute for Computer Sciences,

Social Informatics and Telecommunications Engineeri

ng, vol 536. Springer, Cham. https://doi.org/10.1007/9

78-3-031-48888-7_11

W. Long, S. Li, Y. He, J. Lin, M. Li, and Z. Wen,

"Unraveling Structural Alerts in Marketed Drugs for

Improving Adverse Outcome Pathway Framework of

Drug-Induced QT Prolongation," International Journal

of Molecular Sciences, vol. 24, no. 7, pp. 6771, 2023,

doi: 10.3390/ijms24076771.

Y. Zhou, S. Li, Y. Zhao, M. Guo, Y. Liu, M. Li, and Z.

Wen, "Quantitative Structure–Activity Relationship

(QSAR) Model for the Severity Prediction of Drug-

Induced Rhabdomyolysis by Using Random Forest,"

Chemical Research in Toxicology, vol. 34, pp. 514–

521, 2021, doi: 10.1021/acs.chemrestox.0c00347.

Z. Xiong, Z. Du, X. Rong, Y. Yang, and X. Zhang, "A

Neural Collaborative Filtering Model for Adverse Drug

Reaction Prediction," 2023 2nd International Conferen

ce on Data Analytics, Computing and Artificial Intelli

gence (ICDACAI), pp. 807– 816, doi: 10.1109/ICDA

CAI59742.2023.00158.

ICRDICCT‘25 2025 - INTERNATIONAL CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION,

COMMUNICATION, AND COMPUTING TECHNOLOGIES

720