Researchers from the University of Bern and the National Center of Competence in Research (NCCR) PlanetS have created a machine-learning tool designed to forecast possible planetary configurations featuring Earth-like worlds. This innovation holds the promise to dramatically speed up and thereby transform the ongoing quest for habitable exoplanets across the cosmos.
The quest for finding Earth-like exoplanets—worlds circling stars beyond our solar system—is a key focus in contemporary planetary studies, as these locations are where we might most plausibly discover alien life. Scientists from the University of Bern have recently created a cutting-edge machine learning tool designed to pinpoint planetary systems with potential Earth-like planets.
The whole team involved in this research was associated with the University of Bern and were members of the NCCR PlanetS during their investigation. Dr. Jeanne Davoult, listed as the primary contributor, currently works as a postdoc researcher at DLR (German Aerospace Center) in Berlin. Her work primarily focuses on exoplanet populations, and she crafted the model for her doctorate within the Space Research and Planetary Sciences Division (WP), which falls under the Physics Institute at the University of Bern.
Professor Dr. Yann Alibert, who serves as co-director of the Center for Space and Habitat (CSH), along with Romain Eltschiner, a doctoral candidate at the same center, played crucial roles in advancing this research effort. The findings have only recently been released. published in the journal Astronomy & Astrophysics .
Training using data from the Bern Model
A machine learning model serves as a statistical instrument designed to learn from data, enabling it to identify specific patterns and forecast outcomes. As Dr. Davoult clarifies, “The foundation of our model lies in an algorithm crafted by me which has been taught to discern and categorize planetary systems containing Earth-like planets.”
The model expands upon earlier research to deduce a relationship between the existence or lack thereof of an Earth-like planet and the characteristics of its planetary system.
The algorithm underwent training and testing using data derived from what’s known as the Bern Model of Planet Formation and Evolution. According to co-author Dr. Alibert, this model allows for assertions regarding planetary formation, their evolutionary progression, and the kinds of planets that emerge under specific circumstances within a protoplanetary disk.
Since 2003, the Bern Model has undergone continuous development at the University of Bern. As Dr. Alibert elaborates, "The Bern Model stands out globally for integrating an extensive range of interconnected physical phenomena, which makes studies like the present investigation feasible."
The new model has achieved 99% precision.
The algorithm of the new machine learning model was trained and tested using data on synthetic planetary systems from the Bern Model. "The results are impressive: the algorithm achieves precision values of up to 0.99, which means that 99% of the systems identified by the machine learning model have at least one Earth-like planet," says Dr. Davoult.
The model was subsequently tested using actual observed planetary systems. According to Dr. Davoult, “The analysis revealed that 44 of these systems have a high probability of containing previously undiscovered Earth-like planets. Additional research validated the theoretical potential for such planets to exist within these systems.”
A more effective quest for planets suitable for life
In his master's thesis, Eltschinger, who was also a co-author of the study, helped enhance the machine learning model further. This enhancement enabled the model to be applied across an even broader spectrum of situations.
He states, "These findings hold significant importance for the scientific community, especially for upcoming space endeavors like the PLATO project or conceptual missions such as LIFE, all aimed at exploring and understanding smaller, colder celestial bodies."
Utilizing this machine-learning model to more precisely look for Earth-like planets might reduce search durations and boost the quantity of findings. As Dr. Alibert points out, “This represents a crucial advancement in the quest for planets with environments conducive to life and, eventually, in the hunt for life across the cosmos.”
More information: Jeanne Davoult and others, Predicting earth-like planets: A machine learning method, Astronomy & Astrophysics (2025). DOI: 10.1051/0004-6361/202452434
Furnished by the University of Bern
The tale was initially released on Massima . Subscribe to our newsletter For the most recent science and technology news updates.