MMAO-Cls: Feature Selection and Classifier Tuning

In the field of machine learning, one of the most persistent challenges is the optimal selection of features along with the tuning of classifier hyperparameters. This problem, known as model selection in mixed spaces, combines discrete variables (which attributes to include) and continuous ones (algorithm parameters), exponentially increasing computational complexity. To address this, bio-inspired approaches that emulate natural systems of cooperation and competition have emerged. In particular, the Metabolic Multi-Agent Optimizer (MMAO) has shown potential as a global search mechanism. A recent adaptation, called MMAO-Cls, proposes a scheme where each agent simultaneously encodes a binary feature mask and the classifier hyperparameters, while dynamics such as private energy, communal budget, role drift, and life cycle model the balance between accuracy and complexity. This approach introduces regularization based on the overfitting gap between training and validation, as well as an adaptation of the feature budget based on information priorities. Experimental results on seven standard tabular datasets show that MMAO-Cls achieves competitive performance in validation and testing, standing out for obtaining the most compact feature subsets on average, with a mean selection rate of 0.4881. Although the differences compared to methods like RandomSearch, GA-lite, and PSO-lite are not statistically significant, the method demonstrates its practical utility in scenarios where dimensionality reduction is a priority. From a business perspective, these techniques align with the need to create custom applications that efficiently integrate artificial intelligence. At Q2BSTUDIO, as a custom software development company, we understand that model optimization not only improves accuracy but also reduces operational costs and facilitates deployment in cloud environments. We work with AI for businesses that require robust classification and attribute selection solutions, relying on AWS and Azure cloud services, as well as business intelligence tools like Power BI to visualize the impact of decisions. Furthermore, the incorporation of AI agents and cybersecurity techniques ensures that processes are secure and scalable. MMAO-Cls represents an advancement in automating the machine learning pipeline, and its integration into enterprise platforms allows organizations to obtain more parsimonious models without sacrificing performance. The key lies in combining intelligent exploration of the search space with rigorous complexity control, something that, at Q2BSTUDIO, we apply in business intelligence services and process automation projects. Ultimately, feature selection and classifier tuning are critical processes that, optimized with techniques like MMAO-Cls, can make a difference in the effective adoption of artificial intelligence in the corporate environment.