Symbolic Regression (SR) is a machine learning approach that automatically discovers mathematical expressions to model data. Unlike traditional regression methods, which assume a fixed functional form (e.g., linear or polynomial), SR searches the space of possible equations to find both the structure and parameters that best explain the data. This makes it particularly useful for uncovering interpretable models and hidden relationships in complex systems.
Our Research on Symbolic Regression
Over the last years, our group has explored several directions in Symbolic Regression, with a focus on both methodological advances and practical applications. We have developed a novel multi-objective framework for the generation of interpretable scoring systems [1][2], and investigated survival analysis by combining Cox’s regression with genetic programming [3]. Beyond methodology, we have applied Symbolic Regression to different healthcare problems, showing its potential to uncover meaningful patterns in medical data [4][5]. More recently, we introduced a federated learning framework that enables SR to be performed across distributed datasets without compromising privacy [to appear in ICDM 2025, preprint][6]. Alongside these efforts, we are also pursuing ongoing work on model selection criteria, Bayesian approaches, and other strategies aimed at making Symbolic Regression more robust, efficient, and effective.
You can find the preprint of our latest publication on Federated Symbolic Regression here.
We are always open to collaboration, so if our work resonates with your interests, feel free to get in touch with us.