AI-BASED DIABETES PREDICTION SYSTEM

Volume-09 Issue 10, October-2025 ISSN: 2456-9348 Impact Factor: 8.232 International Journal of Engineering Technology Research & Management (IJETRM) https://ijetrm.com/ IJETRM (http://ijetrm.com/) [408] AI-BASED DIABETES PREDICTION SYSTEM Prof. Aziz Makandar Professor, Department of Computer Science, Karnataka State Akkamahadevi Women’s University, Vijayapura-586101 Mandakini Gangadhar Student, Department of Computer Science, Karnataka State Akkamahadevi Women’s University, Vijayapura-586101 [email protected] Adiba Maniyar Research Scholar, Department of Computer Science, Karnataka State Akkamahadevi Women’s University, Vijayapura-586101 adibamaniy[email protected]m ABSTRACT "AI-Based Diabetes Prediction System," a research work that describes the creation of a novel web application for diabetes risk assessment and early identification. In order to enable prompt intervention, the system looks at a wide range of patient health data in order to identify those who may already have diabetes or are at risk of getting it. A number of machine learning algorithms were applied to bulk amount of data. The various algorithms are Logistic regression, Decision Tree, Random Forest, SVM and ANN are adopted in the study to build an efficient model. The Artificial Neural Network (ANN) classifier with accuracy 85%, precision 84%, recalls 86%, f1-score 85% and AUC 0.89. The model has ability in predicting the patient is diabetic or non-diabetic. A web application that allows user to enter physiological information in real time and dynamically view prediction results was created to improve accessibility and interpretability. The system integrates SHAP-based explainability and feature importance visualizations, providing transparency in model decisions and enabling healthcare professionals to better understand contributing factors. Keywords: Artificial Intelligence, SVM, Random Forest, Artificial Neural Network INTRODUCTION Diabetes Mellitus, or simply diabetes, is one of the most common chronic diseases in the world, affecting people of all ages. The World Health Organisation (WHO) estimates that in 2021, about 537 million adults between the ages of 20 and 79 had diabetes; by 2030, that figure is expected to rise to 643 million and by 2045, it will reach 783 million. Type 1 diabetes, an autoimmune disease in which the pancreas produces little or no insulin, Type 2 diabetes, the most prevalent kind brought on by insulin resistance and lifestyle factors, and gestational diabetes, which develops during pregnancy, are three main categories of diabetes. Since diabetes can cause serious health problems like cardiovascular illnesses, renal failure, blindness, and amputations if left untreated, early detection and prediction are essential. Clinical assays such as the Oral Glucose Tolerance Test (OGTT), Haemoglobin A1C (HbA1c) values, and Fasting Blood Sugar (FBS) are used in traditional diabetes diagnostic procedures. Even though these tests work well, they are frequently laborious, necessitate lab space, and typically only identify the illness after it has advanced significantly. As medical data becomes more widely available, methods for machine learning (ML) and artificial intelligence (AI) have become effective tools for evaluating patient data and identifying those who are at risk of getting diabetes. Healthcare systems may transition to personalised preventive care and early diagnosis by utilising these data-driven strategies. The growing worldwide health burden of diabetes and the necessity for effective predictive tools to support healthcare providers are the driving forces behind this study. Complex patterns in patient health data, such as age, Volume-09 Issue 10, October-2025 ISSN: 2456-9348 Impact Factor: 8.232 International Journal of Engineering Technology Research & Management (IJETRM) https://ijetrm.com/ IJETRM (http://ijetrm.com/) [409] BMI, blood pressure, insulin, glucose levels, and other characteristics that conventional statistical models can find difficult to understand, might be found by AI-based systems. These technologies improve diagnostic accuracy and allow for prompt intervention by acting as assistive decision-support tools for physicians. Furthermore, they empower patients by educating them about their risk factors and encouraging lifestyle changes that can delay the onset of disease. The objective of proposed work is to create an AI-based diabetes prediction system based on patient health metrics and evaluate effectiveness of several machine learning algorithms, including Logistic Regression [7], Decision Tree [8], Random Forest [9] Support Vector Machine (SVM) [10] and Artificial Neural Networks (ANN)[11]. The main goal of proposed work is to determine the best diabetes prediction model and provide an intuitive user interface that enables patients or medical professionals to enter health information and receive prediction results. The approach is appropriate for clinical usage and research since it also produces a structured report that includes the likelihood of diabetes and contributing risk variables. The primary focus of current diagnostic methods is on glucose levels, which restricts their capacity to identify diabetes early. While classic regression-based methods have poor predictive potential, manual diagnosis ignores other significant features. There are wearable glucose monitoring devices available, but they are expensive and intrusive. The suggested system takes into account a number of physiological characteristics, including age, blood pressure, skin thickness, insulin, BMI, diabetes pedigree function, glucose level, and pregnancy, in order to overcome these restrictions. To learn from past trends and make more accurate predictions about future risk, machine learning and deep learning models are trained on datasets like the Pima Indians Diabetes Dataset. Compared to current approaches, the suggested strategy has a number of advantages. It provides probabilistic risk assessment, takes into account a number of risk factors beyond glucose, and obtains a greater accuracy rate of roughly 82% to 85%. The system is also appropriate for real-world healthcare settings since it can be incorporated with mobile health applications, telemedicine platforms, and hospital databases. Reducing the amount of time needed for diabetes screening, increasing the accuracy of early diagnosis with AI, and providing physicians with an intelligent prediction tool are all anticipated results of this study. By warning people about their possible risk of developing diabetes, the system also encourages preventative healthcare, which advances AI-driven medical research and raises the standard of patient care overall. RELATEDWORK Research on diabetes detection and prediction has grown significantly at the nexus of artificial intelligence (AI) and healthcare. While laboratory testing and physician evaluation are the mainstays of traditional clinical diagnosis methods, the rise of Electronic Health Records (EHRs) and publicly accessible datasets has made Machine Learning (ML) and Deep Learning (DL) models very effective in predicting the risk of diabetes [1,8]. H Naz et al [1] have stated that - The study's conclusion demonstrates that DL offers the most promising extracted features and the greatest results. With a 98.07% accuracy rate, DL can be utilised to advance the automatic prognosis tool. By using the omics data for illness onset prediction, the DL approach's accuracy can be further improved. Akihiro Nomura et al [2] have focussed on - Numerous AI/ML-based medical devices have previously received approval from the US Food and Drug Administration in the areas of diabetes, including automatic retinal screening, clinical diagnosis support, and patient self-management tools. As of right now, ML techniques for predicting new-onset diabetes do not outperform traditional risk stratification models that rely on statistical techniques. Mukesh Kumar et al [3] have stated thatThe significance of multilevel ensemble learning is amply demonstrated by the proposed work. Several implemented models were even present, as can be observed, and the process's missing phases resulted in a lack of accuracy. Some authors proceeded with the filthy data after neglecting the pre-processing stages. Authors have typically employed restricted classification techniques. A hybridised model that covers all the necessary processes is presented by the suggested model. A multilevel predictive model is constructed that has a high degree of accuracy—nearly 88.30%—and generalises to predict all class levels. I Tasin et al [4] have stated thatDiabetes may contribute to a lower life expectancy and quality of life. Long-term Volume-09 Issue 10, October-2025 ISSN: 2456-9348 Impact Factor: 8.232 International Journal of Engineering Technology Research & Management (IJETRM) https://ijetrm.com/ IJETRM (http://ijetrm.com/) [410] risk and complications of many diseases can be decreased by early detection of this chronic ailment. This research proposes an autonomous diabetes prediction system that uses a variety of machine learning techniques. This work has made use of a private dataset of female patients from Bangladesh as well as the open-source Pima Indian. Unbalanced class concerns have been addressed through the use of SMOTE and ADASYN preprocessing approaches. Several performance indicators, including precision, recall, accuracy, F1 score, and AUC, were presented in this study for a variety of machine learning and ensemble approaches. Using the ADASYN method, the XGBoost classifier performed the best, with 81% accuracy with an F1 score and an AUC of 0.81 and 0.84, respectively. Varun Jaiswal et al [5] have stated that-This study's goal is to give a general overview of the many machine learning approaches that can be used to automatically predict diabetes. This study describes various machine learning and data mining categorisation methods that have been developed recently for accurate and efficient diabetes diagnosis. Therefore, different methods provide varying levels of accuracy for different types of data. The primary goal of developing diabetes prediction models has changed from increasing forecast accuracy to achieving greater reliability for use in a worldwide population. A small number of techniques that are trained and evaluated on various datasets have been created. Since diabetes is a global issue, a technique that can be used to predict diabetes using a dataset that includes the entire world's population is required. It is anticipated that the present discussion and the suggested framework of algorithms will assist the researcher in creating better prediction models and algorithms to overcome diabetes through early detection. Hafsa Binte Kibria et al [6] have stated thatBased on a number of machine learning techniques, this study successfully predicted the risk of diabetes by utilising a weighted soft voting classifier, two ML algorithms (RF, XGB), and the cross-validation methodology to create an accurate and highly explicable ensemble model. It was shown that the weighted ensemble's predictions outperform those of the individual algorithms by a significant margin. By choosing the right weights, the system was able to attain the best accuracy. The ensemble model outperformed the other models suggested in the literature with an accuracy of 90% and an F1 score of 89%. PROPOSED METHODOLOGY Dataset Description: The dataset is acquired from UCI Machine Learning Repository. It comprises of 8 characteristics (pregnancy, glucose, blood pressure, BMI, age, insulin, etc.) and 768 samples make up the most used benchmark dataset, the Pima Indians Diabetes Dataset (PIDD). The result is either diabetic or non-diabetic. System Architecture: The proposed architecture is described below in figure 1. For the CSV file, the pre-processing is formed. Later split the dataset into train as 80% and test as 20%. Perform training and testing to CSV file to achieve better accuracy. The models performance is evaluated on metrics. Finally, the model has achieved high accuracy and has better prediction system. The working of proposed work is summarized in the form of steps below. The workflow of the proposed study is shown below in figure 1. Figure 1: Proposed work flow Step 1: Data Collection The dataset Pima Indians Diabetes Dataset (PIDD) is acquired from UCI Machine Learning Repository. The data is in CSV file. Step 2: Pre-processing The normalization technique is applied to CSV file and further pre-processing is performed. Volume-09 Issue 10, October-2025 ISSN: 2456-9348 Impact Factor: 8.232 International Journal of Engineering Technology Research & Management (IJETRM) https://ijetrm.com/ IJETRM (http://ijetrm.com/) [411] Step 3: Feature extraction Feature extraction process is performed by adding new features or transforms existing ones depending on medical logics and relationships among data. Step 4: Train Data Initially, split the data in ratio 80:20 for train 80 and for test 20. Later, perform the training and testing to CSV file Step 5: Classifier In Step 5, feed the data to various machine learning classifiers such as Logistic regression, Decision Tree, Random Forest, SVM and ANN. Obtain the accuracy, precision, recall, f1-score, AUC from the proposed model. Step 6: Prediction system Finally, the model is able to predict whether the patient is either diabetic or non-diabetic. Machine Learning Techniques: Logistic Regression: Logistic regression [7] is used to determine the outcome of a categorical dependent variable. As a result, the output needs to be discrete or categorical. It provides probabilistic values that fall between 0 and 1. Decision Tree: A decision tree classifier [8] can be used to tackle problems with regression and classification known as a "tree structured classifier" because each leaf node indicates the classification result, while the interior nodes stand for features Random Forest: The Random Forest classifier [9] is thought of as a regression and classification problem-based classifier. Using the majority vote for classification and the average for regression, it builds decision trees for a variety of datasets. Support Vector Machine (SVM): Support vector machine [10], is a supervised machine learning algorithm used for classification and regression tasks. Artificial Neural Network (ANN) : ANN stands for Artificial Neural Network [11], a computational model inspired by the human brain that is used in machine learning for tasks like pattern recognition and prediction. It consists of interconnected layers of artificial neurons that process information and learn from data. Experimental Result The proposed work is implemented using python programming language on VS Code platform. The operating system is Windows 11 with an Intel Core i5 processor and 8GB of RAM. The various python libraries used are seaborn, matplotlib, tensorflow, keras and numpy. Model Performance: Five distinct algorithms were implemented and compared in order to assess the efficacy of different machine learning models for diabetes prediction: Logistic Regression, Decision Tree Random Forest, Support Vector Machine (SVM) with RBF kernel and Artificial Neural Network (ANN) with two hidden layers. Among all machine learning classifiers, Artificial Neural Network (ANN) has obtained better results with accuracy 85%, precision84%, recall 86%,F1 score 85% and AUC 0.89. Table1: Classification Results of our proposed work Model Accuracy Precision Recall F1-Score AUC Logistic Regression 77% 75% 79% 77% 0.78 Decision Tree 72% 70% 73% 71% 0.74 Random Forest 82% 81% 83% 82% 0.86 SVM (RBF Kernel) 80% 79% 80% 79% 0.84 ANN (2 Hidden Layers) 85% 84% 86% 85% 0.89 The figure 2 shows bar graph representation of all machine learning classifiers in terms of accuracy, precision, recall, f1-score and AUC. Volume-09 Issue 10, October-2025 ISSN: 2456-9348 Impact Factor: 8.232 International Journal of Engineering Technology Research & Management (IJETRM) https://ijetrm.com/ IJETRM (http://ijetrm.com/) [412] Figure 2: Bar Graph representation of classification results Confusion matrix: The confusion matrix showcases the classification results in terms of correctly and wrongly predicted cases. The effectiveness of the suggested machine learning model was assessed. It forecasts the actual condition of patient either diabetic or non diabetic. The confusion matrix of proposed work is shown below in figure 3. Figure 3: Confusion Matrix of proposed work A web application is created for PIMA Indians Diabetes dataset. In order to get a real-time diabetes risk prediction, users can enter important physiological data such as age, BMI, insulin, glucose, and pregnancy. It allows nontechnical users like patients, healthcare, professionals to communicate directly with the model. This interface also improves usability. The web application comprises of Input Parameters Panel and Model Performance Dashboard. Input Parameters Panel: In this web application, the users can enter input values for the above features through control panel. Every feature in CSV file plays a vital task in determining risk of diabetes. Model Performance Dashboard: The web application has a feature of dashboard display. The dashboard display the performance metrics of model which includes Accuracy, F1 Score, Recall, Precision, and AUC . Volume-09 Issue 10, October-2025 ISSN: 2456-9348 Impact Factor: 8.232 International Journal of Engineering Technology Research & Management (IJETRM) https://ijetrm.com/ IJETRM (http://ijetrm.com/) [413] Additionally, SHAP (SHapley Additive explanations) visualizations improve the interpretability of the model: • The SHAP Waterfall Plot shows the positive or negative contributions of specific parameters (such as age, BMI, and glucose) to the prediction outcome. • The SHAP Force Plot provides transparency and confidence in the AI-based diagnosis by dynamically displaying how feature interactions influence the model's judgement. • Glucose is the most important element, followed by BMI and age, according to the Permutation Importance Graph, which ranks features according to their effect on model performance. These visualization tools improve interpretability in machine learning-driven healthcare applications by giving users visibility into both the prediction outcome and the logic behind it. This combination of web technology and data science shows how predictive analytics can be successfully converted into an easy-to-use diagnostic tool that supports diabetes risk assessment early identification and individualized medical insights. The figure 4a,4b,4c and 4d below displays the snapshots of web application which includes inserting input parameters, SHAP (Shapley Additive exPlanations) visualization for improving the interpretability of the model and displaying models performance. Figure 4a: Diabetes prediction web page Volume-09 Issue 10, October-2025 ISSN: 2456-9348 Impact Factor: 8.232 International Journal of Engineering Technology Research & Management (IJETRM) https://ijetrm.com/ IJETRM (http://ijetrm.com/) [414] Figure 4b: Individual Feature Impact on diabetes Figure 4c: visualization of positive and negative feature contribution Volume-09 Issue 10, October-2025 ISSN: 2456-9348 Impact Factor: 8.232 International Journal of Engineering Technology Research & Management (IJETRM) https://ijetrm.com/ IJETRM (http://ijetrm.com/) [415] Figure 4d: Web page illustrating overall model performance Conclusion Artificial intelligence and machine learning can be combined to enhance diabetes risk assessment and early detection, as the AI-Based Diabetes Prediction System. The main aim of proposed study is to build a predictive model that can evaluate patient health data (such as blood pressure, insulin, age, glucose, and BMI) and determine if a person is at risk of getting diabetes. The experimental work is performed on Pima Indians Diabetes Dataset (PIDD). The data is in CSV file. The various machine learning classifies were applied to CSV file and obtained better accuracy in shorter time. The proposed work has attained better results for Artificial Neural Network (ANN) classifier with accuracy 85%, precision 84%, recall 86%, f1-score 85% and AUC 0.89. The model was successful in predicting whether the patient is diabetic or non-diabetic. A web application was created that lets users enter physiological information in real time and see dynamic prediction results. Healthcare professionals can better comprehend contributing elements and model decisions are transparent because to the system's integration of SHAPbased explainability and feature importance visualisations. This interactive method shows how predictive modelling and web deployment can be combined to provide individualised healthcare support and early diabetes risk assessment. References 1) Naz, H., & Ahuja, S. (2020). Deep learning approach for diabetes prediction using PIMA Indian dataset. Journal of Diabetes & Metabolic Disorders, 19(1), 391-403. 2) Nomura, A., Noguchi, M., Kometani, M., Furukawa, K., & Yoneda, T. (2021). Artificial intelligence in current diabetes management and prediction. Current Diabetes Reports, 21(12), 61. 3) Kumar, M., Bajaj, K., Sharma, B., & Narang, S. (2022). A comparative performance assessment of optimized multilevel ensemble learning model with existing classifier models. Big Data, 10(5), 371-387. 4) Tasin, I., Nabil, T. U., Islam, S., & Khan, R. (2023). Diabetes prediction using machine learning and explainable AI techniques. Healthcare technology letters, 10(1-2), 1-10. 5) Jaiswal, V., Negi, A., & Pal, T. (2021). A review on current advances in machine learning based diabetes prediction. Primary Care Diabetes, 15(3), 435-443. 6) Kibria, H. B., Nahiduzzaman, M., Goni, M. O. F., Ahsan, M., & Haider, J. (2022). An ensemble approach for the prediction of diabetes mellitus using a soft voting classifier with an explainable AI. Sensors, 22(19), Volume-09 Issue 10, October-2025 ISSN: 2456-9348 Impact Factor: 8.232 International Journal of Engineering Technology Research & Management (IJETRM) https://ijetrm.com/ IJETRM (http://ijetrm.com/) [416] 7268. 7) Rajendra, P., & Latifi, S. (2021). Prediction of diabetes using logistic regression and ensemble techniques. Computer Methods and Programs in Biomedicine Update, 1, 100032. 8) Sisodia, D., & Sisodia, D. S. (2018). Prediction of diabetes using classification algorithms. Procedia computer science, 132, 1578-1585. 9) VijiyaKumar, K., Lavanya, B., Nirmala, I., & Caroline, S. S. (2019, March). Random forest algorithm for the prediction of diabetes. In 2019 IEEE international conference on system, computation, automation and networking (ICSCAN) (pp. 1-5). IEEE. 10) Joshi, T. N., &Chawan, P. M. (2018). Logistic regression and svm based diabetes prediction system. International Journal For Technological Research In Engineering, 5, 4347-4350. 11) El_Jerjawi, N. S., & Abu-Naser, S. S. (2018). Diabetes prediction using artificial neural network. 12) Kavakiotis, I., Tsave, O., Salifoglou, A., Maglaveras, N., Vlahavas, I., & Chouvarda, I. (2017). Machine Learning and Data Mining Methods in Diabetes Research.Computational and Structural Biotechnology Journal, 15, 104–116. 13) Kumari, V., & Chitra, R. (2013). Classification of Diabetes Disease Using Support Vector Machine. IJERA, 3(2), 1797–1801. 14) Sisodia, D., & Sisodia, D. S. (2018). Prediction of Diabetes Using Classification Algorithms.Procedia Computer Science, 132, 1578–1585. 15) Ali, L., et al. (2019). An Intelligent Healthcare Monitoring Framework Using Wearable SensorsandMachine Learning.IEEE Access, 7, 136552–136563. 16) Zhang, X., et al. (2020). Deep Learning-Based Multimodal Data Integration for Diabetes Prediction.IEEE Journal of Biomedical and Health Informatics, 24(10), 2886–2893. 17) Sun, J., et al. (2021). Time-Series Deep Learning for Blood Glucose Prediction.ScientificReports, 11(1), 19989. 18) Centers for Disease Control and Prevention (CDC). (2020). National Health and Nutrition Examination Survey (NHANES). 19) Kumar, S., et al. (2022). Recent Applications of ML and DL Models in Diabetes: A Comprehensive Review.Diabetology & Metabolic Syndrome, 14(1), 45. 20) Gupta, R., et al. (2024). Hybrid Deep Learning Model for Early Diabetes Risk Prediction.Journal of Supercomputing, 80(6), 5432–5451. 21) Patel, A., et al. (2025). Explainable Deep Learning Framework for Diabetes Mellitus Prediction.Applied Sciences, 15(16), 9162. 22) Khan, M. A., et al. (2024). Feature-Selected Deep Neural Networks for Diabetes Prediction.Journal of Computational Biology and Informatics, 3(2), 15–26. 23) Hasan, F., et al. (2025). Diabetes Prediction from Clinical Data Using Deep Convolution Neural Network.Mathematical Modelling of Engineering Problems, 12(8), 3059–3069. 24) Chen, Y., et al. (2025). Interpretable Machine Learning for Pre-Diabetes Prediction Using CHNS Data.BMC Public Health, 25(1), 22419. 25) Singh, N., et al. (2025). Data Augmentation Approach for Type-2 Diabetes Prediction and Classification.Indian Journal of Science and Technology, 18(4), 529–537.