Nowadays, due to ubiquitous health IoT and digital healthcare, enormous biomedical datasets have been generated. In the face of substantial digital information, an urgent challenge is how to create good data for machine-learning algorithms. Biomedical datasets have the characteristics of high-dimensionality, different sizes, data noises, missing values, and imbalanced data. These complex raw data demote the performance of the machine-learning algorithms in terms of reducing their accuracy, causing overfitting, and taking more time to develop the model. One of the biggest challenges in analyzing biomedical data for critical applications is the complexity of designing an accurate and reliable decision-making algorithm, which avoids overfitting in both small/medium-sized high-dimensional imbalanced datasets. To tackle this challenge, we propose a generalized wrapper-based feature and instance selection, called GeFICA, which is based on two meta-heuristic optimization approaches: a parallel new intelligent genetic algorithm (GA) and a parallel imperialist competitive algorithm (ICA). Our proposed solution is divided into two major categories: First, we present two efficient parallel meta-heuristic optimization methods to solve both discrete and continuous complex optimization problems. Second, using our proposed optimization methods, we propose a feature and instance selection algorithm to improve the high dimensionality and imbalance problems in medical datasets using the proposed optimization methods. Our proposed framework significantly improve the accuracy and efficiency of high-dimensional imbalanced numeric datasets under different sizes with try to avoid of overfitting.