Models and Dataset

<b>P3DE (Parameter-less Population Pyramid with Deep Ensemble):</b><br>P3DE is a hybrid feature selection framework that combines the Parameter-less Population Pyramid (P3) metaheuristic optimization algorithm with a deep ensemble of autoencoders. Designed for high-dimensional biological data, P3DE dynamically evaluates candidate feature subsets using an ensemble of autoencoders with different activation functions (Sigmoid, Tanh, ReLU). Ensemble weights are computed based on initial, historical, and functional reconstruction performance. This parameter-free, adaptive architecture enhances exploration and avoids overfitting, making P3DE a robust tool for biomarker discovery in gene expression datasets.<br><b>TJO (Tom and Jerry Optimization):</b><br>TJO is a nature-inspired metaheuristic algorithm that models the predator-prey dynamics of the cartoon characters Tom (predator) and Jerry (prey). Operating in a binary search space, TJO simulates intelligent and evasive movements of the prey to guide the population toward optimal solutions. The algorithm does not rely on predefined control parameters like crossover or mutation rates, which makes it lightweight and easy to implement for various feature selection and optimization tasks.<br><b>RAO (Rao Optimization Algorithm):</b><br>RAO is a parameter-less optimization algorithm that updates solutions based on simple arithmetic operations involving the best and worst individuals in the population. Unlike conventional evolutionary algorithms, RAO does not use mechanisms such as crossover, mutation, or selection. Its simplicity and lack of algorithm-specific parameters make it computationally efficient and easy to apply in high-dimensional problems such as gene selection for cancer classification.