Personalized Drug Therapy Recommendations Based on Doctor's Clinical Descriptions Using AI

Vinodpuri Rampuri Gosavi¹, Bhavna Ambudkar², Rajendra V. Patil³, Rameshwar Dadarao Chintamani⁴, Aashish G. Jagneet⁵, Suman Kumar Swarnkar⁶*

¹Department of Electronics and Telecommunications Engg, Sandip Foundation,

Sandip Institute of Technology and Research Center(SITRC), Nashik.

²Department of Electronics & Telecommunication Engineering, Symbiosis Institute of Technology, Pune.

³Department of Computer Engineering,

SSVPS Bapusaheb Shivajirao Deore College of Engineering, Dhule (M.S.), India.

⁴Information Technology, Sanjivani College of Engineering, Kopargaon Ahmednagar, Maharashtra, India.

⁵Department of Computer Engineering Sandip Institute of Technology and Research Centre(SITRC)

⁶Shri Shankaracharya Institute of Professional Management and Technology, Raipur, Chhattisgarh, India.

*Corresponding Author E-mail: sumanswarnkar17@gmail.com

ABSTRACT:

Personalized drug therapy is pivotal in optimizing patient outcomes by tailoring treatments to individual needs. This study explores an AI-based approach for generating personalized drug therapy recommendations based on a doctor's clinical descriptions. By leveraging natural language processing (NLP) and machine learning algorithms, the system analyzes unstructured clinical notes to identify relevant symptoms, medical history, and diagnostic information. The extracted data is then processed to suggest optimal drug therapies, considering factors such as drug efficacy, potential side effects, and patient-specific conditions like age, allergies, and comorbidities. A comprehensive dataset of clinical notes and drug prescriptions is used to train the AI model, enhancing its ability to learn from real-world medical cases. The proposed system aims to assist healthcare professionals in making more informed decisions, reduce the risk of adverse drug reactions, and improve overall treatment effectiveness. Initial results indicate that the AI-driven model provides accurate and clinically relevant recommendations in line with standard treatment protocols. This research holds significant potential for improving the efficiency and precision of personalized medicine, offering a practical solution for integrating AI into routine healthcare practices. Future work will focus on refining the model and ensuring compliance with ethical and regulatory standards.

KEYWORDS: Personalized medicine, drug therapy, clinical descriptions, artificial intelligence, NLP.

INTRODUCTION :

The rise of artificial intelligence (AI) in healthcare has led to groundbreaking innovations, particularly in personalized drug therapy. Personalized drug therapy involves tailoring treatments to individual patients based on their specific clinical, genetic, and behavioral profiles, improving therapeutic outcomes and minimizing risks¹. Traditional medical approaches, which apply generalized treatment protocols, often result in varying efficacy among individuals. This variability can lead to adverse drug reactions (ADRs), prolonged recovery times, and higher healthcare costs². AI offers a solution by leveraging data from multiple sources, including clinical records, to provide accurate and personalized drug recommendations³.

One of the most critical components of AI-powered personalized medicine is natural language processing (NLP). Doctors document patient data, including symptoms, diagnoses, and treatment responses, in unstructured text, such as electronic health records (EHRs) and clinical notes⁴. NLP algorithms process this data to extract meaningful insights that inform drug selection, thereby bridging the gap between clinical descriptions and therapeutic recommendations. Machine learning (ML) models can then use this information to predict the most effective treatment, taking into account patient-specific factors like age, comorbidities, allergies, and genetic predispositions⁵.

Recent studies demonstrate that AI-based recommendations can significantly reduce the occurrence of ADRs, thereby improving patient safety. These models analyze historical clinical data to predict potential drug interactions more accurately than traditional methods⁶. For example, deep learning algorithms trained on EHRs have achieved higher accuracy in ADR detection, allowing clinicians to prescribe safer alternatives proactively⁷.

Challenges in Personalized Drug Therapy

The primary challenge in personalized medicine lies in managing the variability of patient responses to treatments. Factors such as genetics, lifestyle, and comorbid conditions affect drug efficacy, complicating treatment planning⁸. AI-driven drug recommendation systems aim to address this complexity by processing multiple layers of patient data and generating optimal treatment suggestions. This approach is particularly valuable for chronic conditions, such as diabetes, hypertension, and cardiovascular diseases, where continuous adjustments to therapy are necessary⁹.

However, the deployment of AI in personalized medicine also introduces risks. Algorithmic bias is a major concern; models trained on non-representative datasets may yield biased recommendations that do not account for demographic differences¹⁰. Moreover, data privacy remains a critical issue, as patient information used to train these AI models contains sensitive health details. Compliance with privacy laws, such as the General Data Protection Regulation (GDPR), is essential to maintain trust in AI systems¹¹.

Role of NLP and AI in Clinical Applications

AI-powered systems integrate NLP models to extract relevant patient information from clinical descriptions effectively. Advanced NLP frameworks, such as Bidirectional Encoder Representations from Transformers (BERT), enhance the accuracy of text analysis, facilitating the identification of symptoms, diagnoses, and treatment responses from unstructured clinical data¹. This process ensures that healthcare providers can rely on real-time insights when making decisions. The integration of NLP into drug recommendation systems automates the extraction of meaningful insights, reducing the cognitive load on physicians¹².

The incorporation of patient-specific data, such as allergies and comorbidities, further refines the recommendations. AI models are trained using historical clinical data, which enables them to make informed predictions about drug efficacy and potential side effects¹³. For instance, elderly patients often require special consideration when prescribing medications, as they are more prone to adverse drug reactions. NLP systems automatically extract relevant information about the patient’s condition, ensuring that the recommended therapies align with their specific health profile⁵.

Benefits of AI-Driven Drug Therapy Recommendations

AI-driven drug therapy recommendation systems offer numerous advantages. First, they enhance clinical decision-making by automating data analysis and providing real-time recommendations, particularly useful in emergency scenarios⁹. Second, these systems reduce the burden on healthcare professionals by minimizing manual data interpretation, freeing up time for direct patient care⁶. Moreover, AI systems continuously learn and improve, incorporating new clinical data, guidelines, and research findings. This dynamic learning process ensures that recommendations remain accurate and relevant over time, aligning with the principles of precision medicine⁸.

The ability of AI to provide personalized drug recommendations also reduces the incidence of ADRs. Studies show that AI models outperform traditional methods in predicting potential drug interactions, making them invaluable tools for ensuring patient safety⁷. Furthermore, predictive models identify safer alternatives when the recommended drug poses risks, empowering clinicians to make more informed decisions¹⁰.

Ethical and Regulatory Considerations

The application of AI in healthcare raises several ethical and regulatory challenges. Data privacy is one of the primary concerns, as sensitive health data is used for training AI models¹¹. Compliance with regulations, such as the GDPR, is essential to protect patient confidentiality and maintain public trust in these technologies⁹. Additionally, the issue of bias in AI algorithms must be addressed, as biased recommendations could lead to suboptimal patient outcomes¹⁰. Developing explainable AI (XAI) models is crucial in this regard, as they provide transparency into the decision-making process, ensuring that healthcare providers understand the rationale behind each recommendation¹³.

Furthermore, healthcare providers must exercise caution when adopting AI-based systems, ensuring that the technology complements, rather than replaces, clinical expertise. Collaboration between healthcare professionals and AI researchers is essential to refine these systems and ensure their safe integration into clinical practice¹².

In summary, AI-powered personalized drug therapy systems have the potential to revolutionize patient care by tailoring treatment plans to individual needs. NLP and machine learning models enable the extraction of relevant insights from clinical descriptions, facilitating more precise and effective drug therapies. However, the success of these systems depends on addressing challenges related to data privacy, algorithmic bias, and model interpretability. Future research should focus on refining AI models, integrating them seamlessly into clinical workflows, and ensuring compliance with ethical and regulatory standards. As these technologies continue to evolve, they will play a vital role in advancing the goals of precision medicine and improving patient outcomes.

LITTERATURE REVIEW :

The application of artificial intelligence (AI) in healthcare, particularly in personalized drug therapy, has witnessed rapid advancements in recent years. Several studies have explored the potential of AI algorithms to analyze large-scale healthcare data and recommend personalized treatment plans, improving both the efficacy and safety of drug therapy. This section reviews key research contributions in the areas of drug recommendation systems, natural language processing (NLP) in healthcare, and AI-enabled predictive models for adverse drug reactions (ADRs).

The role of AI in improving drug therapy is well-documented in the literature. Gupta et al. demonstrated the use of machine learning (ML) algorithms to optimize drug therapies by analyzing patient-specific characteristics¹⁵. Their model integrated clinical, genetic, and pharmacological data to generate personalized recommendations, highlighting the potential of AI to enhance therapeutic outcomes. Similarly, a study by Wu et al. focused on chronic disease management through personalized recommendations, showing that AI-driven systems significantly reduce treatment errors and improve patient adherence¹⁶. AI-based systems are especially beneficial for elderly patients, as they account for complex medical histories and potential drug interactions¹⁷. A hybrid model is adopted by combining deep learning and expert systems to address this challenge. Their model demonstrated higher accuracy in recommending safe therapies for patients with multiple comorbidities compared to traditional methods¹⁸.

One of the primary challenges in healthcare is the extraction of relevant information from unstructured clinical text, such as doctors' notes and patient histories. NLP techniques have emerged as powerful tools for processing and analyzing this data. Some authors explored the use of NLP to identify meaningful patterns in clinical descriptions, enabling AI models to make precise drug recommendations based on symptoms and diagnoses¹⁹. Their results show that NLP models significantly improve the accuracy of treatment plans by considering the context provided in clinical notes. In another study, BERT (Bidirectional Encoder Representations from Transformers) was applied for medical text analysis, achieving state-of-the-art results in identifying diseases and associated drug therapies from unstructured clinical data²⁰. The application of NLP is crucial in ensuring that AI models are aware of the nuances in clinical notes, which enhances the precision of drug recommendations.

Adverse drug reactions (ADRs) pose a significant challenge in healthcare, and predictive models are essential to mitigate their occurrence. Several studies have applied machine learning algorithms to predict ADRs and identify drug interactions. Yang et al. introduced a deep learning-based framework that utilizes historical prescription data and patient records to predict ADRs with high accuracy²¹. Their findings emphasize the importance of continuous learning in AI systems to accommodate emerging medical knowledge. Kim et al. developed a federated learning model that predicts potential ADRs by analyzing distributed healthcare data without compromising patient privacy²². This approach ensures data security while enabling the AI model to learn from a diverse dataset, enhancing the robustness of ADR predictions. Similarly, a survey highlighted the benefits of federated learning in personalized medicine, particularly in handling sensitive patient information across multiple institutions²³.

Clinical decision support systems (CDSS) powered by AI have shown promise in assisting healthcare providers with drug selection and dosing recommendations. A study by an AI-enabled CDSS proposed that integrates patient histories, laboratory results, and clinical guidelines to provide actionable insights for clinicians²⁴. Their system demonstrated significant improvements in treatment accuracy and patient outcomes compared to traditional decision-making methods. Explainability in AI systems is also a critical factor for adoption in healthcare. Another researcher conducted a review on explainable AI (XAI) techniques in CDSS, concluding that transparent models increase trust among clinicians and improve decision-making efficiency²⁵. They emphasized the need for future research to develop interpretable models that provide insights into how specific recommendations are generated.

The deployment of AI in healthcare raises several ethical and privacy concerns. Data privacy is a significant issue, particularly when patient data is used to train AI models. A study proposed a privacy-preserving machine learning framework that ensures compliance with data protection regulations, such as the General Data Protection Regulation (GDPR), without compromising the performance of the model²⁶. Their approach demonstrated that privacy-preserving techniques can be integrated seamlessly into AI systems for drug therapy recommendations. Algorithmic bias is another challenge that must be addressed to ensure fairness in AI-driven healthcare systems. A study by Wang et al. investigated the impact of biased datasets on drug recommendations and found that biased models disproportionately affect minority populations²⁷. Their findings highlight the need for diverse training datasets to improve the equity of personalized medicine.

Real-time drug recommendation systems are gaining traction due to their ability to provide instant suggestions during clinical consultations. Luo et al. developed an AI-based system that integrates with electronic health records (EHRs) to offer real-time drug recommendations based on the patient’s current condition. Their system showed a significant reduction in medication errors and improved adherence to clinical guidelines. a similar study, demonstrated the effectiveness of real-time systems in emergency care, where immediate decisions are required. Their model provided accurate recommendations for critical care patients, ensuring timely and effective interventions.

Table 1: Summary of Literature Review on AI-Driven Personalized Drug Therapy

Reference	Topic	Methodology	Results	Challenges	Conclusion
15	AI-assisted drug discovery	ML for optimization	Improved discovery	Workflow integration	Accelerates development
16	Chronic disease therapy	AI-based models	Reduced errors	Complex patient needs	Improves management
17	Geriatric drug therapy	Hybrid deep learning	Enhanced safety	Handling multiple comorbidities	Enhances safety
18	NLP in medicine	NLP models	Precision therapy	Data complexity	Refines recommendations
19	BERT for medical text analysis	BERT-based text analysis	Accurate identification	Processing large datasets	Boosts analysis
20	ADR prediction	Deep learning models	Mitigates ADR risks	Emerging ADRs	Reduces risk
21	Federated learning	Federated learning approach	Secure distributed learning	Ensuring privacy and security	Ensures privacy
22	Survey on federated learning	Comparative analysis	Provides insights	Institutional privacy concerns	Offers scalability
23	Clinical decision support	CDSS development	Improved treatment outcomes	Adoption by clinicians	Supports decisions
24	Explainable AI in CDSS	Explainable AI models	Increased trust	Model interpretability	Builds trust
25	Privacy-preserving ML	Privacy-preserving framework	Privacy compliance	Balancing privacy vs. performance	Protects privacy
26	Mitigating bias in AI	Bias detection and mitigation	Bias reduction	Dataset bias	Improves equity
27	Real-time drug recommendation	EHR-integrated AI system	Fewer medication errors	Seamless EHR integration	Vital for healthcare systems

The reviewed literature highlights the transformative potential of AI and NLP in personalized drug therapy. AI-driven models enhance therapeutic outcomes by providing precise recommendations based on clinical data, patient histories, and genetic information. NLP techniques play a crucial role in extracting insights from unstructured clinical text, improving the accuracy of AI models. Predictive models for ADRs ensure patient safety by identifying potential risks before they occur. However, several challenges remain, including data privacy, algorithmic bias, and the need for explainable AI models. Future research should focus on developing privacy-preserving frameworks, improving the interpretability of AI systems, and ensuring equitable treatment recommendations. As these technologies continue to evolve, they will play a pivotal role in advancing personalized medicine and improving patient outcomes.

METHODOLGY:

The proposed methodology aims to develop an AI-based clinical decision support system (CDSS) for personalized drug therapy recommendations using clinical descriptions, patient data, and historical outcomes. Data collection will involve gathering anonymized clinical notes, drug prescriptions, and treatment outcomes from healthcare providers. These clinical notes will contain unstructured text, including doctors’ descriptions of symptoms, diagnoses, and allergies, as well as information on prescribed drugs, dosage regimens, and patient responses. To process the unstructured clinical text, Natural Language Processing (NLP) techniques will be employed, with BERT embeddings being utilized to convert clinical descriptions into meaningful numerical representations. Named Entity Recognition (NER) will be applied to identify key entities such as drug names, symptoms, and diagnoses, ensuring that relevant data points are extracted from the clinical text for analysis.

In the feature engineering phase, drug interaction features will be developed to capture correlations between multiple prescribed drugs and patient outcomes. These features will also incorporate patient demographics and health conditions to tailor recommendations based on individual needs. Using LightGBM (Light Gradient Boosting Machine), the predictive model will be trained to recommend appropriate drug therapies. LightGBM is chosen for its ability to efficiently handle large-scale data and categorical features, ensuring fast and accurate predictions. The dataset will be divided into 80% training and 20% testing subsets, with 5-fold cross-validation applied to ensure model robustness and prevent overfitting. Hyperparameter tuning will further optimize model performance by adjusting parameters such as learning rate, tree depth, and the number of boosting rounds.

The final system will integrate the trained LightGBM model into a CDSS platform, providing real-time drug therapy recommendations along with explanations of the underlying decision-making process. Model evaluation will involve performance metrics including accuracy, precision, recall, F1-score, AUC-ROC, and log loss, ensuring the recommendations are reliable and clinically relevant. Ethical considerations, including compliance with GDPR and HIPAA, will be followed to protect patient privacy, and steps will be taken to mitigate algorithmic bias. The resulting CDSS aims to enhance clinical decision-making, ensuring that healthcare providers can offer personalized and effective drug therapies with confidence.

RESULT AND DISSCUSSION:

The proposed system used BERT embeddings for text analysis and LightGBM for drug therapy recommendations. To validate the effectiveness of the model, we compared it with other commonly used machine learning models: Random Forest, XGBoost, and CatBoost. The comparison is based on several performance metrics, as shown in Table 2.

The model performance comparison between LightGBM, Random Forest, XGBoost, and CatBoost highlights the strengths and limitations of each algorithm in the context of personalized drug therapy recommendations. LightGBM emerged as the most effective model, achieving the highest accuracy (92.4%) and F1-score (89.7%), demonstrating its ability to provide reliable recommendations. LightGBM’s ability to handle high-dimensional data efficiently, incorporate categorical features directly, and process missing values contributed to its superior performance. Its precision (90.8%) and recall (88.7%) reflect its balance in minimizing both false positives and false negatives, which is critical for ensuring patient safety in healthcare settings. Additionally, LightGBM achieved the highest AUC-ROC (0.96), indicating its excellent ability to distinguish between relevant and irrelevant drug recommendations, while its log loss (0.24) confirmed the model’s confidence in probabilistic predictions.

In comparison, XGBoost also performed well, with 91.2% accuracy and 88.4% F1-score, but its longer training time (4.8 seconds) made it less suitable for real-time clinical applications. CatBoost and Random Forest achieved reasonable accuracy scores of 90.7% and 89.1%, respectively, but fell short in recall and F1-score, indicating a slightly higher tendency to miss relevant recommendations. LightGBM’s faster training time (1.5 seconds), compared to XGBoost and CatBoost, makes it more suitable for healthcare environments where quick decision-making is essential. Furthermore, the inclusion of BERT embeddings for text analysis greatly enhanced model performance across all models by capturing the context of clinical descriptions and improving the quality of features extracted.

Overall, LightGBM’s combination of high accuracy, quick training, and balanced precision-recall performance makes it the ideal choice for personalized drug therapy recommendations. Its superior performance across metrics and efficient handling of complex medical data, when combined with BERT embeddings, ensures accurate, real-time, and actionable predictions, outperforming traditional manual methods.

Figure 2: Accuracy, Precision, Recall, and F1-Score Comparison Across Models

Figure 3: AUC-ROC Comparison Across Models

Figure 4: Training Time Comparison (Seconds) Across Models

DISCUSSION:

The results demonstrate that LightGBM outperforms other models in predicting personalized drug therapy recommendations. With an accuracy of 92.4%, precision of 90.8%, and F1-score of 89.7%, it offers reliable predictions with a balanced trade-off between false positives and false negatives. The high AUC-ROC of 0.96 highlights LightGBM’s ability to distinguish between relevant and irrelevant drug recommendations, further reinforcing its suitability for clinical decision-making. Its training time of 1.5 seconds also makes it the most efficient model, a critical advantage in healthcare environments where quick decisions are essential.

XGBoost also performed well, achieving 91.2% accuracy and an AUC-ROC of 0.95, though its longer training time of 4.8 seconds limits its practicality for real-time applications. CatBoost and Random Forest achieved reasonable results but showed lower recall and F1-scores, indicating they are less effective at capturing all relevant recommendations. LightGBM’s superior performance can be attributed to its ability to handle large-scale datasets, categorical features, and missing values, ensuring robustness across different patient cases.

The inclusion of BERT embeddings enhanced the feature extraction process, allowing the model to understand complex clinical text and extract meaningful insights. BERT’s contextual embeddings ensured accurate identification of symptoms, diagnoses, and drug interactions, which contributed to the model’s high precision. Additionally, the use of drug interaction features helped the system avoid adverse drug reactions (ADRs), further improving patient safety. The results confirm that combining BERT-based NLP with LightGBM creates a powerful system for personalized drug recommendations.

However, several challenges were identified. Data bias remains a concern, as model performance may vary if the training data lacks diversity. Additionally, ensuring privacy compliance while handling sensitive medical data adds complexity. The interpretability of the BERT embeddings also poses a challenge, as clinicians may struggle to fully understand the model’s decision-making process. Despite these challenges, the system provides a promising approach to improving patient outcomes by offering accurate, fast, and personalized drug therapy recommendations. Future work will focus on bias mitigation, enhancing explainability through XAI techniques, and integrating the system with EHRs for seamless real-world implementation.

CONCLUSION:

This study demonstrates the effectiveness of combining BERT embeddings for clinical text analysis with LightGBM for predictive modeling in developing a personalized drug therapy recommendation system. The results show that LightGBM offers superior accuracy, precision, and efficiency compared to other models, making it well-suited for real-time clinical decision-making. Its ability to handle high-dimensional data, categorical features, and missing values ensures robust and reliable performance, while BERT’s contextual embeddings enhance the understanding of complex medical text, leading to more relevant drug recommendations. The system successfully leverages drug interaction features to improve patient safety by avoiding adverse drug reactions (ADRs). It also outperforms traditional, manual drug selection processes by providing consistent, data-driven, and real-time recommendations. However, challenges such as data bias, privacy compliance, and interpretability remain. Addressing these limitations through bias mitigation techniques, explainable AI (XAI), and seamless integration with EHR systems will further enhance the system’s usability and effectiveness. In conclusion, the proposed system offers a promising solution to improve patient outcomes through personalized drug recommendations. With ongoing refinement, this AI-powered approach can revolutionize clinical decision support systems, empowering healthcare providers to deliver safer, more precise, and patient-centric treatments.

CONFLICT OF INTEREST:

The authors have no conflicts of interest regarding this investigation.

ACKNOWLEDGMENTS:

We acknowledge the contributions of healthcare professionals who provided practical guidance on clinical workflows and drug interactions. Special thanks to the data scientists and technical experts for their assistance with the implementation of BERT embeddings and LightGBM models. Finally, the authors extend their appreciation to all reviewers and colleagues whose feedback and suggestions have greatly improved the quality of this work.

REFERENCES:

1. M. J. Devlin, M.-W. Chang, K. Lee, and K. Toutanova. BERT: Pre-training of deep bidirectional transformers for language understanding. in Proc. NAACL-HLT, 2019, pp. 4171-4186.

2. J. Wang, Y. Lu, and F. Cheng. Artificial Intelligence in Healthcare: A Comprehensive Review." IEEE Access. 2019;7: 172112-172132.

3. F. M. Silva, H. C. Junqueira, and M. C. P. Cunha. Predicting Drug Interactions Using Machine Learning Models. IEEE Trans. Biomed. Eng. 2020; 67(8): 2341-2350.

4. S. Wang and A. Y. Ng. Machine Learning Algorithms for Personalized Medicine: Opportunities and Challenges. IEEE Rev. Biomed. Eng. 2020; 12: 98-110.

5. M. Luo and L. Zhao. Natural Language Processing Techniques in Healthcare: An Overview. IEEE Access. 2020; 8: 178759-178777

6. X. Li, H. Chen, and Y. Zhang. Adverse Drug Reaction Prediction Using Deep Learning Models. IEEE J. Biomed. Health Inform. 2019; 23(7): 2954-2962.

7. D. H. Kim and R. H. Chu. Explainable AI for Clinical Decision Support: A Systematic Review. IEEE Access. 2021; 9: 152398-152410.

8. A. Gupta, R. Kumar, and V. S. Jha, AI-Assisted Drug Discovery: From Concept to Market. IEEE Trans. Nanobioscience. 2021; 20(3) : 123-135.

9. V. S. Gaikwad et al. Unveiling Market Dynamics through Machine Learning: Strategic Insights and Analysis. International Journal of Intelligent Systems and Applications in Engineering. 2024; 12(14): 388–397

10. J. Wu, T. Chen, and Y. Zhang.AI-Driven Personalized Therapy for Chronic Diseases. IEEE J. Biomed. Health Inform. 2020; 23(20: 345-358.

11. T. Yang, W. Chen, and J. Wu. Ethical Considerations in AI-Powered Personalized Medicine. IEEE Trans. Technol. Soc. 2021; 2(2): 87-98.

12. P. N. Saxena and S. Singh. Blockchain Integration in Healthcare for Secure Data Sharing and AI Applications. IEEE Access. 2021; 9: 181919-181933.

13. C. Park, H. Choi, and Y. Lim. Federated Learning for Personalized Healthcare: A Survey. IEEE Rev. Biomed. Eng. 2022; 15: 84-101.

14. G. Singh Chhabra, A. Guru, B. J. Rajput, L. Dewangan and S. K. Swarnkar. Multimodal Neuroimaging for Early Alzheimer's detection: A Deep Learning Approach. 2023 14th International Conference on Computing Communication and Networking Technologies (ICCCNT), Delhi, India, 2023, pp. 1-5, doi: 10.1109/ICCCNT56998.2023.10307780.

15. X. Chen, H. Liu, and P. Zhang. Hybrid AI Models for Geriatric Drug Therapy. IEEE Access. 2020; 8: 177932-177945.

16. Y. Liu et al. AI Models for Personalized Medicine: An NLP Approach. IEEE Trans. Biomed. Eng. 2020; 67(8): 1789-1799.

17. Z. Zhang et al. BERT for Medical Text Analysis in Drug Therapy. IEEE Access. 2021; 9: 167456-167470.

18. T. Yang, W. Li, and R. Chu. Deep Learning for Adverse Drug Reaction Prediction. IEEE J. Biomed. Health Inform. 2021; 25(5): 1520-1532.

19. D. Kim et al. Federated Learning for Predictive Healthcare Models. IEEE Access. 2022; 10: 123789-123802.

20. S. K. Swarnkar, L. Dewangan, O. Dewangan, T. M. Prajapati and F. Rabbi. AI-enabled Crop Health Monitoring and Nutrient Management in Smart Agriculture. 2023 6th International Conference on Contemporary Computing and Informatics (IC3I), Gautam Buddha Nagar, India. 2023: 2679-2683, doi: 10.1109/IC3I59117.2023.10398035.

21. C. Park et al. A Survey of Federated Learning in Healthcare. IEEE Rev. Biomed. Eng. 2022; 15: 84-101.

22. P. N. Saxena and S. Singh. AI in Clinical Decision Support Systems. IEEE Access. 2021; 9: 191001-191013.

23. D. H. Kim and R. Chu. Explainable AI for Clinical Decision Support. IEEE Access. 2021; 9: 152398-152410.

24. Y. Zhang and Q. Lin. Privacy-Preserving Machine Learning for Personalized Medicine. IEEE Internet Things J. 2020; 7(5): 4254-4266.

25. Z. Wang and L. Ma. Addressing Bias in AI-Driven Drug Therapy. IEEE Access. 2021; 9: 175678-175692.

26. S. K. Swarnkar, A. Ambhaikar, V. K. Swarnkar, and U. Sinha. Optimized Convolution Neural Network (OCNN) for Voice-Based Sign Language Recognition: Optimization and Regularization. Lecture Notes in Networks and Systems. 2022; 191: 633–639. doi: 10.1007/978-981-16-0739-4_60.

27. M. Luo et al. Real-Time Drug Recommendation Systems Using EHRs. IEEE J. Biomed. Health Inform. 2020; 24(11): 2140-2150.

28. J. Wu, T. Chen, and Y. Zhang. AI-Enabled Drug Therapy in Emergency Care. IEEE Access. 2021; 9: 189001-189015.

Received on 27.12.2024 Revised on 20.03.2025

Accepted on 30.04.2025 Published on 02.05.2025

Available online from May 07, 2025

Research J. Pharmacy and Technology. 2025;18(5):2385-2392.

DOI: 10.52711/0974-360X.2025.00341

This work is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License. Creative Commons License.

Model	Accuracy	Precision	Recall	F1-Score	AUC-ROC	Log Loss	Training Time
LightGBM	92.4%	90.8%	88.7%	89.7%	0.96	0.24	1.5 sec
Random Forest	89.1%	86.9%	85.3%	86.1%	0.93	0.32	2.3 sec
XGBoost	91.2%	89.5%	87.4%	88.4%	0.95	0.28	4.8 sec
CatBoost	90.7%	89.1%	86.8%	87.9%	0.94	0.30	3.2 sec