Bio-Inspired Optimisation and Deep Learning Approaches for Diabetes Prediction: A Comprehensive Survey

Abstract

Diabetes mellitus, a chronic metabolic disorder, poses a growing global health challenge requiring timely and accurate diagnosis. Traditional diagnostic methods often fall short due to their manual and error-prone nature. Recent advancements in machine learning (ML), deep learning (DL), and bio-inspired optimisation have significantly enhanced the potential for early diabetes prediction. This survey presents a comprehensive overview of state-of-the-art techniques integrating ML/DL models with nature-inspired algorithms for improved diagnostic accuracy and feature selection. It examines traditional classifiers, ensemble methods, and deep architectures like CNNs and LSTMs, highlighting their suitability for complex clinical data. Bio-inspired algorithms such as Genetic Algorithm (GA), Particle Swarm Optimisation (PSO), and Shuffled Frog Leaping Algorithm (SFLA) are discussed for their effectiveness in dimensionality reduction and model tuning.

Additionally, the paper covers essential preprocessing steps, including handling missing data, outliers, class imbalance, and feature scaling. Explainable AI (XAI) methods, evaluation metrics, and cross-validation strategies are explored to ensure model transparency and reliability. The survey also identifies emerging trends such as federated learning and Bayesian deep learning, outlining future directions for building trustworthy, scalable, and clinically applicable diabetes prediction systems.