The Minimum Description Length Principle algorithm (MDLP) is a widely used supervised discretization technique, although its classical implementation may prove inefficient when applied to large volumes of data. In this work, we propose an improved version of MDLP that incorporates the following strategies aimed at reducing execution time without compromising preprocessing quality: enforcing a minimum partition size, limiting recursion depth, randomly subsampling candidate points, using cumulative counts for efficient statistical computation, and parallelizing the process running the algorithm. The proposed approach was evaluated using datasets from the UCI repository, comparing its performance against a classical implementation in R using Random Forest for the classification task of the discretized sets. The results show that the improved MDLP achieves faster execution times in the discretization process while maintaining accuracy in the classification task. This implementation represents an efficient and scalable alternative for preprocessing continuous data in supervised learning contexts.