Hyperglycemia Prediction Using Machine Learning

Early diagnosis is crucial for controlling and preventing long-term problems of diabetes, which affects a vast number of people globally. To predict diabetes using the K-Nearest Neighbour (KNN) algorithm, which is a popular machine learning approach. Initially, we performed data preprocessing by eliminating missing values and outliers. We used several techniques, such as correlation analysis and recursive feature elimination, for feature selection. Following that, we divided the dataset into separate sets for training and testing purposes. We have integrated several machine learning methods, including logistic regression, decision trees, random forests, and support vector machines, to predict the onset of diabetes. Utilized a publicly available dataset containing patient information such as age, gender, BMI, blood pressure, and glucose levels, which has been preprocessed by removing missing values and outliers. The KNN algorithm is efficient to the pre-processed dataset to guess the probability of the patient having diabetes created on their features. The study assesses the effectiveness of the KNN algorithm using various performance metrics. Additionally, the elbow method is utilized to determine the optimal value of K. Furthermore, it includes a comparison of KNN’s performance than decision tree and logistic regression. The research concludes that KNN algorithm provides satisfactory accuracy in predicting diabetes, and the optimal value of K is found to be 7 using the elbow method. Furthermore, KNN algorithm is found to outperform such as decision tree and logistic regression.

I. Introduction

This goal is to create a precise predictive model for recognizing individuals at risk of rising diabetes. Early diagnosis and management of diabetes can significantly improve patient’s health outcomes and reduce the risk of complications. Machine learning has been an active tool for medical research and diagnosis, including diabetes research. Machine learning algorithms can analyze immense datasets of patient information to recognize patterns and associations that may not be closely visible to humans [1]. We have used machine learning algorithms to examine large datasets of patient information, including demographic data, medical history, lifestyle factors, and clinical pointers such as blood glucose levels. The model will be advanced using supervised learning techniques, where the algorithm is trained on a considered dataset of diabetes patients and non-diabetes patients. We have incorporated several steps, including data collection and preprocessing, feature production, algorithm selection, model training, testing, and evaluation. The data will be sourced from electronic health records and other medical databases, preprocessed, and relevant features selected [2]. The chosen algorithm will be trained on a labeled dataset and evaluated on a separate testing dataset. The predictive model can help healthcare workers identify patients at high risk of developing diabetes, leading to timely intrusions and management to prevent complications and improve health outcomes. To inform public health policies and interventions aimed at preventing diabetes, potentially reducing healthcare cost and improving productivity.