Diabetes is a growing health problem affecting millions of people worldwide. If not diagnosed early, it can lead to serious complications such as heart disease, kidney failure, and vision loss. Current methods for diagnosing diabetes rely on medical tests, which can be expensive, time-consuming, and sometimes unavailable in certain areas. Our research focuses on using artificial intelligence (AI) and machine learning to develop a faster, more accurate, and cost-effective way to predict diabetes. By using patient data such as glucose levels, body mass index (BMI), and insulin levels, we can create smart models that help doctors identify high-risk patients before symptoms appear.
One of the biggest challenges in using AI for medical diagnosis is ensuring that the system focuses on the most important data. Our research solves this problem by using advanced selection techniques to identify the key factors that contribute to diabetes. Instead of analyzing unnecessary information, our method chooses only the most relevant features, improving accuracy and efficiency. We also fine-tuned our model’s settings to make sure it works as effectively as possible. In testing, our model outperformed many traditional methods, making it a promising tool for early diagnosis.
Another major issue in AI-based diagnosis is that medical data is often unbalanced. In most diabetes datasets, there are far more records of healthy individuals than diabetic patients, which can lead to inaccurate predictions. To overcome this, we applied data balancing techniques that ensure the model learns from both groups equally. This means our system is better at detecting real diabetes cases, reducing the chance of misdiagnosis and making it a more reliable tool for doctors and patients.
We tested our model using 768 patient samples, where 268 samples were diabetic and 500 were non-diabetic. Our approach achieved an impressive accuracy of 97.5%, significantly higher than many existing methods. When compared to previous studies, our model outperformed them by margins ranging from 6.5% to 19.64%. These results show that our system is not just an improvement—it could be a real game-changer for diabetes prediction.
The impact of our research goes beyond just building a better prediction model. By identifying the key risk factors for diabetes, our work can help healthcare providers develop targeted prevention strategies. If doctors know what early signs to look for, they can recommend lifestyle changes and treatments before the disease worsens, which could save lives and ease the pressure on healthcare systems.
Looking ahead, we are proud to share that we are already supported by the Dean of Scientific Research and the AAU Incubator. With their help, we are taking the first steps to turn our project into a real-world product. We also plan to reach out to local incubators and organizations to secure additional support and funding. Our goal is to develop a simple digital tool that people can use through smartphones or local clinics, especially helping those in remote areas. By building this complete project, we hope to make diabetes prediction more accessible and affordable for everyone.
