MUMBAI, India, Aug. 31 -- Intellectual Property India has published a patent application (202641101124 A) filed by Vignans Nirula Institute Of Technology And Science For Women on August 21, 2026, for A Hybrid Optimized Ensemble System And Method For Early Detection Of Anemia Using Supervised Machine Learning Algorithms.
Inventors include Sk. Sharmila; G. Rohini Phaneendra Kumari; Jagadish Kalava; G. Kusuma Harinadh; B. Manasa; B. Jahnavi Krishna; G. Akanksha; and G. Bhavya Reddy.
The application for the patent was published on August 28, 2026, under issue no. 35/2026.
Abstract: The present invention discloses a Hybrid Optimized Ensemble Anemia Detection (HOEAD) system and method for the early, accurate, and interpretable detection and severity stratification of anemia from hematological and demographic patient data, overcoming the accuracy-interpretability trade-off and the static, single-algorithm limitations of existing anemia-prediction systems. The invention combines an Optimized Random Forest classification engine, whose hyperparameters are automatically tuned by grid- search cross-validation, with a Hybrid Support Vector Ensemble classification engine that employs a weighted combination of linear, polynomial, and radial basis function kernels to capture both linear and nonlinear relationships among hematological indices such as haemoglobin concentration, haematocrit, mean corpuscular volume, mean corpuscular haemoglobin, and red cell distribution width. An adaptive ensemble fusion engine dynamically weights the outputs of the two constituent classification engines according to their cross-validated accuracy, convergence stability, and historical reliability, so as to generate a final anemia risk prediction that consistently outperforms either constituent classifier alone. Moreover, the severity stratification module stratifies a positive classification result according to its level of severity—namely, mild, moderate, or severe, and also determines the nature of the physiological type of the anemia condition, while the explainability module highlights the contributions of each hematological factor to the classification result in order of magnitude. The continuous learning module iteratively tunes the constituent classification engines and their weights based on new clinical data with labels, which come periodically over time, without retraining the entire model from scratch. The experimental results show that the proposed HOEAD model obtains an accuracy of 97.8%, a precision of 96.5%, a recall of 98.9%, an F1 score of 97.7%, and an ROC-AUC of 98.2%, and outperforms the Gradient Boosting Machines, Neural Networks, and Naive Bayes baseline models in terms of interpretability and computational efficiency.
Disclaimer: Curated by HT Syndication.
