Diabetes is a chronic illness that interferes with your body\'s ability to control blood sugar or glucose levels. Our body\' cells depend on glucose as a major energy source, and the hormone insulin, which the pancreas produces, aids in controlling glucose levels. This research suggests a machine learning based diabetes detection system. The method makes use of a dataset of patient records that detail several clinical traits and whether diabetes is present or not. The most crucial characteristics for predicting diabetes are determined via a feature selection technique. The dataset is then used to train and evaluate machine learning techniques such as Random Forest, Support Vector Machine, KNN and Logistic Regression. The model’s performance can be evaluated using certain metrics. The outcomes demonstrate that the suggested system beats the numerous baseline models and has good predictive accuracies for diabetes using Random Forest algorithm. This approach may be helpful for detecting diabetes in early stage and can help in improving the outcome of patients.
Diabetes is a chronic illness that affects millions of individuals worldwide and, if ignored, is linked to serious health issues. Diabetes must be identified and managed early in order to reduce these consequences and enhance patient outcomes. In the field of healthcare, notably in the prediction and diagnosis of diseases, machine learning approaches, particularly the random forest algorithm, have shown considerable promise. In this study, we suggest a random forest-based diabetes diagnosis method. The method makes use of a dataset of patient records that include a number of clinical characteristics, including age, BMI, blood pressure, and the presence or absence of diabetes. To determine which traits are most crucial for predicting diabetes, we use a feature selection process. The dataset is then used to train and test a random forest model. The random forest algorithm is an ensemble technique that brings together various decision trees to increase prediction accuracy. A randomly chosen subset of features and a randomly chosen sample of training data are used to construct each decision tree in the random forest. Combining all of the decision trees' projections yields the ultimate conclusion. The suggested random forest algorithm-based diabetes diagnosis system has a number of advantages over conventional statistical approaches. It can handle highly dimensional, non-linear data and record intricate relationships between various aspects. A personalized approach to interventions and treatments may be made possible by the system's insights on the risk factors for diabetes. The system's effectiveness is assessed using a number of parameters, including accuracy, precision, recall, and F1 score. The outcomes show that the suggested method beats numerous baseline models and achieves high accuracy in predicting diabetes. The random forest model's most critical properties for predicting diabetes, according to the feature importance analysis, are age, BMI, and blood pressure.
II. LITERATURE REVIEW
K.VijiyaKumar et al. provides us the information about the Diabetes disease and explains to us how the rapid growth of the disease is affecting the people. Also they have assumed that by the year 2035 the patients will be doubled as 592 Million. To detect diabetes in an earlier stage they have used Random Forest Algorithm as it provides more accurate results than other models.
S.Kranthi Reddy et al. have used the PIMA Indian Diabetes Dataset from Kaggle to detect the diabetes in female patients. They have used two algorithms namely Random Forest and K-Nearest Neighbor having accuracy of 78.4% and 80.8% respectively. Where they used K-Fold Cross validation technique to train the dataset applying 11 folds .
L.V.Rajani Kumari et al. provides us information about how diabetes is affecting people and how Machine Learning can be used to predict diabetes in early stages. They have used several Machine Learning models such as K-Nearest Neighbor, Logistic Regression, Naive Bayes and Random Forest having accuracy about 78.57%, 72%, 71% and 76% respectively. By using Evaluation Metrics they have explained which model is best suited for predicting diabetes .
Deeraj Shetty et al. explains how Data mining techniques can be used to extract useful data from the Hospital data. The primary target of this examination is to assemble an Intelligent Diabetes Disease Prediction System that gives analysis of diabetes malady utilizing diabetes patient's databases.
In this system, we propose the use of algorithms like Bayesian and KNN (K-Nearest Neighbor) to apply on a diabetes patient’s database and analyze them by taking various attributes of diabetes for prediction of diabetes disease .
Narendra Mohan et al. explains how support vector machine [SVM] is applied in diabetes prediction. The performance of the SVM algorithm is analyzed for different available kernels. The best kernel is selected and used for prediction. The proposed work is implemented in python programming language and its performance is as good as other algorithms .
III. PROPOSED SYSTEM
A. System Architecture of Diabetes Detection System.
The proposed System Architecture of Diabetes Detection System is described in the figure below:
In this project, the major language used is Python with the help of Jupyter Notebook and Streamlit for creating the front-end. The Oversampling Technique SMOTE is used to balance the dataset while K-Fold Cross Validation Technique is used to train the Dataset. With the application of Random Forest Algorithm the accuracy score of 83% has been achieved. We transformed our model into a pickle model to connect it with the front-end. Hence, we created a web application to detect Diabetes in person using the medical parameters and values.
The first step is to collect the dataset which we did from Kaggle. The name of the dataset is PIMA Indian Diabetes Dataset. This dataset consists of female patients with a minimum age of 21 consisting of their diagnostic parameters. After this step, the next step was data preprocessing which includes data cleaning and removal of null values. Now it is seen that the dataset has 768 values which includes 500 data points for Non-Diabetic patients and 268 data points for Diabetic patients. So it is evident that the dataset is highly imbalanced which can cause issues for the models that have to be applied. Hence, to solve these issues a balancing technique can be used such as oversampling.
IV. FUTURE SCOPE
Adding New Features: To increase the detection of diabetes, the system can be improved by adding new clinical elements, such as genetic data and lifestyle factors.
Utilizing Variable Technologies: The system can be improved by integrating wearable technology to offer continuous data on patient health, such as that provided by smartwatches and activity trackers.
Real Time Monitoring: Clinical support systems can be integrated into the system to give healthcare professionals real-time forecasts and insights into the variables influencing diabetes risk.
Personalized Interventions: Interventions that are specifically tailored to the needs of each patient can be treated using the system, which will enhance patient outcomes and lower healthcare expenditures.
Extending to Other Diseases: Other diseases can be added to the system to help with early identification and care, such as cancer and cardiovascular disease.
The proposed system is an effective instrument for the early diagnosis and management of diabetes. In terms of forecasting diabetes, the Random Forest algorithm performed better than numerous other baseline models. By implementing more sophisticated machine learning methods and interfacing with the electronic health record systems, the system can be improved. The technology has the potential to fundamentally alter how medical professionals diagnose and manage diabetes, resulting in better patient outcomes and higher quality of life for diabetic patients.
 K.Vijayakumar, B.Lavanya, I.Nirmala, S.Sofia Caroline, “ Random Forest Algorithm for Prediction of Diabetes”, International Conference on Systems Computation and Networking, 2019.
 S Kranthi Reddy, T Krishnaveni, G.Nikitha, E.Vijayakanth, “ Diabetes Prediction Using Different Machine Learning Algorithms ”, Third International Conference on Inventive Research in Computing Applications (ICIRCA-2021).
 Deeraj Shetty, Kishor Rit, Sohail Shaikh, Nikita Patil, “ Diabetes Disease Prediction Using Data Mining ”, 2017 International Conference on Innovations in Information, Embedded and Communication System (ICIIECS).
 Liu Lei, “ Prediction of Score of Diabetes Progression Index Based on Logistic Regression Algorithm”, 2020 International Conference on Virtual Reality and Intelligent Systems (ICVRIS).
 Narendra Mohan, Vinod Jain, “Performance Analysis of Support Vector Machine in Diabetes Prediction”, Fourth International Conference on Electronics, Communication and Aerospace Technology (ICECA-2020).
 Prof.Rajesh Lomte, Sheetal Dagale, Sneha Bhosale, Shraddha Ghodake, “ Survey Of Different Feature Selection Algorithms For Diabetes Mellitus Prediction”, 2018 Fourth International Conference on Computing Communication Control and Automation (ICCUBEA).
 Hala Alshamlan, Hind Bin Taleb, Areej Al Sahow, “ A Gene Prediction Function for Type 2 Diabetes Mellitus using Logistic Regression ”, 2020 11th International Conference on Information and Communication Systems (ICICS).
 Vinod Jain, “Performance Analysis of Supervised Machine Learning Algorithm for Prediction of Diabetes”, International Conference on Edge Computing and Applications (ICECAA 2022).
 Prakhar Saxena, Subhadeep Saha, S. Kiruthika Devi, “ Analysis and Prediction of Diabetes Using Machine Models”, 2022 International Mobile and Embedded Technology Conference (MECON).
 Mrs. L.V Rajani Kumari, P. Shreya, Mehrunnisa Begum, T. Pavan Krishna, M. Prathibha, “ Machine Learning Based Diabetes Detection”, 2021 6th International Conference On Communication And Electronic Systems (ICCES).