AI model developed by OUTR, Saudi univ researchers predicts liver disease with 95.49% accuracy

Bhubaneswar: A new hybrid AI model developed by researchers at Odisha University of Technology and Research (OUTR) here and Prince Sattam bin Abdulaziz University in Saudi Arabia has exhibited high accuracy in predicting liver disease, with a success rate of 95.49%. The model works combines two types of artificial intelligence — deep learning and boosting techniques — to improve the accuracy of results.

Their study, ‘Liver Disease Prediction Using a Hybrid Machine Learning Approach’, published in the Feb issue of the journal, ‘Engineering, Technology & Applied Science Research’, said the tool could make early screening faster, cheaper and easier, especially in areas with a shortage of liver specialists. Liver disease is a major global health concern and is often detected late, making treatment harder. Tests such as biopsy and imaging scans are costly, invasive and require specialised doctors.


The model was trained on the Indian Liver Patient Dataset (ILPD), which has 583 patient records with basic health parameters such as age, bilirubin levels, proteins and liver enzymes. The team also derived additional features, including ratios such as direct-to-total bilirubin, to help the system detect subtle changes linked to liver health.

As the dataset had more liver disease cases than healthy ones, the researchers used SMOTE-Tomek to balance it, improving learning and reducing errors in identifying both categories.

The study reported strong discrimination between healthy and diseased cases. The model achieved a precision of 98.4% and specificity of 98.50%, indicating fewer false positives and high accuracy in identifying those without the disease. Researchers said it can run on standard computers, making it suitable for govt hospitals, small clinics and telemedicine setups.

Sanjit Kumar Dash, an author and faculty member at OUTR, said the model is aimed at supporting early diagnosis, as many liver diseases show no symptoms until they become severe. Early detection can improve treatment outcomes and reduce pressure on the healthcare system.

Mohammed Altaf Ahmed, corresponding author from the Saudi university, said the hybrid approach helps the model capture complex medical patterns while remaining efficient, and can serve as a decision-support tool in settings with limited diagnostic facilities.

The authors cautioned that the reported accuracy is based on controlled experiments using ILPD and real-world performance may vary due to differences in patients, lab conditions and medical practices. They said further validation on larger, multi-hospital datasets is needed before wider clinical use.

“Collaboration between computer scientists, clinicians and policymakers is essential to build reliable AI tools that improve early diagnosis and patient outcomes,” Dash said.