Enhancing Dorsalgia Prediction using Neutrosophic Sets in a Genetic Algorithm-Optimized Hybrid CNN-LSTM Framework on Spinal Geometry Parameters

Abstract

Predicting dorsalgia involves a multifaceted approach that encompasses the analysis of demographic, lifestyle, and medical data. Machine learning algorithms and advanced data analytics play a pivotal role in forecasting the risk of developing back pain. Early prediction aids in proactive interventions and personalized healthcare strategies, thereby mitigating the burden of dorsalgia on individuals and healthcare systems. The proposed feature selection is the initial feature set’s most educational elements by evolutionary gravitational search-based feature selection (EGSFS). Specifically, the framework is trained and fine-tuned using spinal geometry parameters, enabling precise identification of individuals at risk of developing dorsalgia. This study presents a novel approach for classification tasks using a Genetic Algorithm (GA)-optimized hybrid Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) model. The GA optimizes the model’s architecture and hyperparameters to enhance its performance. The framework is implemented using Python. In the categorization procedure, the Single Neutrosophic sets aid in capturing ambiguity, which is particularly beneficial when handling dorsalgia disorders that may present with confusing symptoms, thus enhancing the accuracy of classifying various dorsalgia conditions. Experimental results demonstrate that this hybrid approach significantly improves classification accuracy, making it a viable option for several practical applications. Experimental results exhibit remarkable improvements in accuracy and predictive power, underscoring the potential of this innovative approach in advancing preventative and personalized healthcare strategies for back pain management. The experiment was built on the lower back pain symptoms dataset. A comparison is made between the experimental results and previous prediction models like Logistic Regression, Decision Tree Classifier, Random Forest, and Support Vector Machine in terms of accuracy, F1-score, precision, and recall. The accuracy of normal and abnormal data is 99%.