44 Hidden Earth-Like Planets Identified by AI Models
- Researchers using machine learning have identified 44 planetary systems that potentially harbor undiscovered Earth-like planets, CNBC Indonesia reported based on a study published in the journal Astronomy &...
- Jeanne Davoult developed the machine learning algorithm during her doctoral research at the University of Bern alongside Romain Eltschinger and Yann Alibert.
- The computer simulations incorporated stars with masses equivalent to the Sun, half the Sun, and one-fifth of the Sun.
Researchers using machine learning have identified 44 planetary systems that potentially harbor undiscovered Earth-like planets, CNBC Indonesia reported based on a study published in the journal Astronomy & Astrophysics. Developed by researchers at the University of Bern and the National Centre of Competence in Research Planet in Switzerland, the algorithm achieved a 99 percent precision rate when tested on computer-simulated planetary systems.
Algorithm Design and Computer Simulation
Jeanne Davoult developed the machine learning algorithm during her doctoral research at the University of Bern alongside Romain Eltschinger and Yann Alibert. The model utilizes known planetary arrangements to search for hidden planets by analyzing how mass and orbital distances retain traces of formation processes. Because astronomers rarely observe every planet orbiting a given star, the team relied on the Bern Model of Planet Formation and Evolution to generate tens of thousands of synthetic planetary systems.
The computer simulations incorporated stars with masses equivalent to the Sun, half the Sun, and one-fifth of the Sun. Researchers simulated embryo growth, migration, and gravitational interactions over 20 million years, followed by calculations tracking cooling, contraction, atmospheric loss, and tidal migration up to 10 billion years. Before training the model, the team concealed planets whose gravitational influence fell below specific detection thresholds to mimic the limitations of radial velocity observations.
Precision Rates and Observational Limits
The machine learning model employs a random forest method consisting of 500 decision trees, with each tree voting on whether a system contains an Earth-like planet. Researchers used 80 percent of the synthetic data to train the model and the remaining 20 percent for testing. When the voting threshold was set above 90 percent, the model achieved a 99 percent precision rate on populations with larger stellar masses, while models trained on the smallest stars reached 94 percent precision.
Despite the high precision scores in simulations, the predicted planets remain unconfirmed through telescope observations. Verifying the existence of these candidate worlds depends on future observational campaigns.