Abstract

Liver disease poses a significant global health concern, impacting millions of individuals. Timely and accurate diagnosis is vital for effective disease management and better patient outcomes. Machine learning (ML) techniques have shown great promise in predicting various medical conditions, including liver diseases. However, the success of ML models heavily relies on the quality and quantity of training data. Unfortunately, many datasets suffer from class imbalance, where certain classes, such as diseased and non-diseased patients, are not equally represented. This imbalance can lead to biased predictions and reduced model accuracy, making it crucial to address this issue to enhance the reliability of liver disease prediction using ML models. Therefore, this project aims to overcome the challenge of class imbalance by employing advanced data balancing algorithms. Our proposed system involves preprocessing the dataset using the Synthetic Minority Over-sampling Technique (SMOTE) that generates synthetic samples for the minority class, creating a more balanced dataset. Additionally, it also adjusts the learning process's cost function to account for the class imbalance, further improving the model's performance. Once we have a balanced dataset, an ML model (logistic regression, support vector classifier, and gradient boosting classifier) is trained to predict liver disease. The proposed model is evaluated on an independent test dataset, using various metrics such as accuracy, precision, recall, and F1-score to assess its effectiveness. By effectively handling class imbalance through data balancing algorithms, this model is expected to offer valuable support to medical practitioners in diagnosing liver diseases early and accurately, ultimately leading to improved patient care and outcomes.

Description