Spandana Sujay is a second-year student majoring in Computer Science Engineering at RV University, Bengaluru, India. She has a keen interest in art, music, enjoys exploring new things, and constantly seeking to learn new skills. With a positive attitude and enthusiasm, she is eager to contribute positively to the field and is keen to embrace every opportunity for growth and excellence. Her curiosity and passion drive her to explore diverse areas, including technology and creative arts, as she prepares to make meaningful contributions to her field.

Poster Abstract

Introduction

Diabetes is a major global health issue, leading to many illnesses and deaths worldwide. Predicting who might develop diabetes and how it will progress is essential for better treatment and care decisions. In response to this critical need, the present research proposes an approach that leverages machine learning techniques for diabetes prediction. By analyzing patterns and risk factors associated with diabetes, machine learning models can provide early warnings and personalized insights, enabling more effective management and prevention of this chronic disease. Globally, diabetes affects approximately 8.5% of adults aged 20-79 years and around 20% of individuals aged 80 years and older. These statistics underscore the importance of proactive strategies to mitigate diabetes-related risks and improve public health outcomes.

Objective

The primary objective is to develop predictive models using HPCC Machine Learning bundles, evaluating them based on specific performance metrics to determine the most accurate one. By examining basic medical data, these models aim to accurately predict which patients are at risk of developing diabetes. This approach will help in implementing preventive measures and enhancing treatment effectiveness.

Methodology

In this poster, patient records are analyzed comprehensively to explore the predictive power of various features in diabetes management and prevention. Beyond merely constructing predictive models, the analysis explores into the significance of each feature in predicting diabetes, enhancing our comprehension of the relevance of various characteristics in diabetes management.

Post data collection, the model development process unfolds through several key phases:

Firstly, data preprocessing involved preparing the data for analysis, ensuring it was in a suitable format for modeling.

Secondly, in the model training phase, HPCC Learning Trees models, including the Random Forest algorithm, were utilized for both classification and regression tasks to predict diabetes outcomes using the preprocessed data. This approach aimed to enhance diabetes management and prevention strategies through effective machine learning techniques.

Thirdly, evaluation and optimization steps appraised model performance using specific metrics. Optimization focused on fine-tuning algorithmic hyperparameters and refining data processing methods to attain optimal outcomes.

Poster also has an option to plot graphs to highlight key findings, enhancing the clarity and impact of the results.

This study seeks to advance diabetes prediction through the effective application of machine learning models. In Python, the algorithm achieved a notable 97% accuracy, demonstrating its potential to significantly enhance public health by improving diabetes management and prevention strategies. The analysis of feature importance provides critical insights into the factors influencing diabetes development, offering a solid foundation for future research and more effective preventive measures. Meanwhile, the implementation of the algorithm in ECL is still in progress, with results forthcoming.

Presentation

In this Video Recording, Spandana provides a tour and explanation of her poster content.

Machine Learning Approach to Diabetes Detection:

Click on the poster for a larger image.