The thyroid gland significantly impacts almost every metabolic function in the body through the hormones it generates. Disruption of this gland can lead to thyroid disease, a widely occurring endocrine condition globally. Thyroid diseases can range from a minor, benign goiter needing no intervention to potentially fatal cancer. Existing research applies machine learning classification and optimization techniques but often struggles to explore neural network architectures efficiently, resulting in suboptimal performance and longer training times for thyroid disease classification. This research aims to anticipate hypothyroidism in its early stages using a hybrid framework integrating Genetic Algorithms (GAs) and Monte Carlo Tree Search (MCTS) to optimize a Multi-Layer Perceptron (MLP) for thyroid detection. We leverage GA to generate a diverse population of possible neural network models, each encoded as a chromosome in the genetic tree. The classification accuracy of each model serves as the fitness criterion. MCTS is then used to navigate the genetic tree, exploring promising branches and exploiting high-performing models. The optimal solution from the MCTS–GA–MLP framework is an enhanced MLP that attains the highest classification accuracy for thyroid conditions. This strategy surpasses various other classification algorithms in key metrics. The proposed framework is extendable to other medical datasets and serves as a powerful tool for developing robust, efficient machine learning models tailored to medical diagnostics.