Improved Clinical Diagnosis Using Predictive Analytics

• Published in: IEEE access Vol. 10; pp. 75158 - 75175
• Main Authors: N, Divyashree, K S, Nandini Prasad
• Format: Journal Article
• Language: English
• Published: Piscataway IEEE 2022; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Algorithms; Artificial neural networks; Classification; Classification algorithms; Clinical decisions; Data models; Decision analysis; Decision making; Decision trees; Diagnosis; Feature extraction; LWGMK-NN; machine learning; Mathematical analysis; Performance measurement; Prediction algorithms; Prediction models; Predictive analytics; Random forests; Random sampling; Sampling methods; Statistical analysis; supervised classification; Support vector machines
• ISSN: 2169-3536; 2169-3536
• DOI: 10.1109/ACCESS.2022.3190416

The strength of Predictive analytics lies in the ability to reveal interesting patterns obscured within the data and aid the decision-making process. Predictive analytics when deployed using high-end techniques, tools, and methods plays a significant life savior role in the early detection and diagnosis of several human diseases. This research work proposes the implementation of predictive analytics using a multi-stratified algorithm named the "Local Weight Global Mean K-Nearest Neighbor (LWGMK-NN)" under the supervised classification category built over the foundation of analytical techniques without any preset assumptions to discover insights and make predictions. Ten standard clinical datasets are considered to demonstrate the performance of the proposed work against nine state-of-the-art classification algorithms: Logistic Regression, Decision trees, Gaussian Naive Bayes, Random Forest, Linear Support Vector Machine, Stochastic Gradient Descent, Artificial Neural Networks, and XGBoost used as benchmark algorithms. Experimental results shown through performance metrics obtained for simple random sampling, 5-fold cross-validation- a statistical re-sampling method, and 5 times iterated 5-fold cross-validation techniques justify the efficiency of the LWGMK-NN algorithm, and its implementation as a predictive model.