02245nas a2200205   4500000000100000008004100001653003000042653001200072653000900084653003100093653003200124653001800156100001200174700001200186700001800198245012800216856009500344300001600439520158400455       2024                            d10aElectronic Health Records10aLeprosy10aLSTM10aChicken Swarm Optimization10aParticle Swarm Optimization10aDeep learning1 aMehta J1 aKumar A1 aChakrabarti P00aSeverity Classification of Electronic Health Records of Leprosy Patients Using Chicken Swarm Optimization-based LSTM Model   uhttps://bpasjournals.com/library-science/index.php/journal/article/download/2618/1775/3988  a13155-131693 a<p>Leprosy is a persistent infectious illness produced by bacteria called Mycobacterium leprae. Lepra responses and delayed leprosy diagnosis might result in chronic neuritis and eventually disability. This paper aims to classify leprosy patients' Electronic Health Records (EHRs) based on the severity of their diseases as managed at the designated center in India. The health records include demographic information, signs of symptoms, laboratory test results, and clinical notes. The leprosy EHRs are modeled with Systematized Nomenclature of MedicineClinical Terms (SNOMED-CT) medical codes. To predict the future clinical events deep learning algorithms using Long short-term memory (LSTM) networks have been validated in the last few years. However, the performance of the LSTM network depends upon the selection of hyperparameters. We propose a novel method of optimizing LSTM parameters using chicken swarm optimization (CHSO) to classify the severity levels of leprosy patients as mild, moderate, or severe cases. Instead of using trial and error-based hyperparameter tuning, the proposed CHSO-LSTM model provides the best fittest value of hyperparameters to increase the accuracy of classification. The result shows that our model outperforms ANN, GRU, LSTM, Bi-LSTM, and Particle Swarm Optimization-based LSTM deep learning models concerning performance matrices Accuracy (ACC) - 99.11%, Precision (PRE) - 99%, Recall (REC) - 99%, F1-Score (F1-SCR) - 99%, Matthew’s correlation coefficient (MCC) - 0.9902 and categorical cross-entropy loss (CCE-loss) - 0.0197.</p>