EFFICIENT MACHINE LEARNING FOR SMEs: A LOW-COST BAGGING-BASED METHOD FOR REDUCED DATA TRAINING

Abstract

In the context of industrial automation, small and medium-sized enterprises face major challenges when applying Machine Learning technologies due to the high computational cost, complexity, or the need for large data sets for training the models. This work focuses on the application of REDIBAGG, a Bagging-based ensemble method designed to reduce the size of the training sets while maintaining the predictive performance. By generating reduced bootstrap samples, REDIBAGG decreases execution times and storage requirements, providing a computationally efficient alternative without compromising performance. REDIBAGG was evaluated on industrial data sets with high dimensionality, varying the number of instances and features. It has been tested using common off-the-shelf classifiers without specific tuning, ensuring ease of implementation. The results demonstrate that REDIBAGG significantly reduces training times while preserving classification accuracy comparable to traditional Bagging. These reductions are particularly relevant in resource-constrained industrial environments. The low computational cost of REDIBAGG and its ability to provide good classification position it as an option for small and medium-sized enterprises looking to integrate Machine Learning into their automation processes. This work contributes a scalable approach that lowers the barriers to the adoption of Machine Learning in industrial environments, aligning with the goals of low-cost data analytics and digital transformation initiatives.