Design of artificial neural networks optimized through genetic algorithms and sequential quadratic programming for tuberculosis model

This paper proposes an innovative method for the tuberculosis (TB) model based on a hybrid technique which combines a feed-forward neural network (FFNN) with a Genetic Algorithm (GA) and Sequential Quadratic Programming (SQP) methodologies. The algorithm's main optimizer is GA, while SQP is employed to fine-tune GA's outputs in order to boost assurance in the result. The TB model consists of five classes: susceptible individuals; latent carriers of TB who are unrecognized; individuals with active tuberculosis being treated at home; individuals with active tuberculosis who are being treated at a hospital; and recovered individuals. The nonlinear differential TB system is used to develop a log sigmoid fitness-based function employing mean squared error. The provided paradigm's stability, accuracy, and usefulness are compared using Adam's numerical technique and absolute error analysis. Furthermore, for repeated large algorithm runs, the convergence evaluations of mean absolute deviation (MAD), root mean square error (RMSE), and Theil's inequality coefficient (TIC) is conducted for each class of TB model. The algorithm's precision is demonstrated by the accuracy of convergence measures for MAD, RMSE, and TIC, which range from 3 to 14 decimal places. The value of the proposed approach-based stochastic algorithm is supported by the quantitative study's findings.