Transfer Learning Approach for Learning of Unstructured Data from Structured Data in Medical Domain

Transfer learning is a most important research area within information retrieval. As we know, there are different types of data available everywhere and among those, dealing with unstructured data is quite difficult. This paper focuses on dealing with unstructured data. Social challenge is, any non-medical background person also uses this system for prediction of patient disease. This paper utilizes a bisecting k-means algorithm for the purpose of disease prediction. We have proposed a model for identifying more relevant disease using readings mentioned in patient's pathology lab test report. Our model is influenced by clustering and unsupervised transfer learning. We demonstrate the effectiveness of our model using patient pathology lab report dataset and dataset used for storing different test names (hemoglobin, sugar, etc.) of four diseases (Diabetes, Lipid profile cholesterol, Liver profile and Kidney profile). Our basic aim is to improve performance of the system by transferring knowledge, learned in one or multiple source tasks and use the same to improve learning in a related target task.