Implementing WEKA for medical data classification and early disease prediction

TLDR: This research work comprehensively compared different data classification techniques and their prediction accuracy for chronic kidney disease dataset using performance measures like ROC, kappa statistics, RMSE and MAE using WEKA tool.

Abstract: In recent years, the advent of latest web and data technologies has encouraged massive data growth in almost every sector. Businesses and leading industries are viewing these huge data repositories as a tool to design future strategies, prediction models by analyzing patterns and gaining knowledge from this unstructured data by applying different data mining techniques. Medical domain has now become richer in term of maintaining digital records of patients related to their diagnosis and treatment. These huge data repositories can range from patient personnel data, diagnosis, treatment histories, test diagnosis, images and various scans. This terabytes of medical data is quantity rich but weaker in information in terms of knowledge and robust tools to identify hidden patterns of knowledge specifically in medical sector. Data Mining as a field of research has already well proven capabilities of identifying hidden patterns, analysis and knowledge applied on different research domains, now gaining popularity day by day among researchers and scientist towards generating novel and deep insights of these large biomedical datasets also. Uncovering new biomedical and healthcare related knowledge in order to support clinical decision making, is another dimension of data mining. Through massive literature survey, it is found that early disease prediction is the most demanded area of research in health care sector. As health care domain is bit wider domain and having different disease characteristics, different techniques have their own prediction efficiencies, which can be enhanced and changed in order to get into most optimize way. In this research work, authors have comprehensively compared different data classification techniques and their prediction accuracy for chronic kidney disease. Authors have compared J48, Naive Bayes, Random Forest, SVM and k-NN classifiers using performance measures like ROC, kappa statistics, RMSE and MAE using WEKA tool. Authors have also compared these classifiers on various accuracy measures like TP rate, FP rate, precision, recall and f-measure by implementing on WEKA. Experimental result shows that random forest classifier has better classification accuracy over others for chronic kidney disease dataset.