Astronomers are increasingly turning to artificial intelligence to help process the massive datasets generated by NASA’s space telescopes, with recent efforts yielding dozens of newly validated exoplanets from observations that might otherwise have been overlooked.
A new study demonstrates how machine learning algorithms can help distinguish genuine planetary signals from false positives in data from NASA’s Transiting Exoplanet Survey Satellite (TESS). The research, published in Monthly Notices of the Royal Astronomical Society, describes an AI-assisted validation process that confirmed 120 exoplanet candidates from TESS observations.
The work addresses one of the most persistent challenges in exoplanet detection: determining whether a dimming signal from a distant star indicates an orbiting planet or results from other astrophysical phenomena, such as eclipsing binary star systems or stellar variability.
“The main challenge is that there are many astrophysical scenarios that can mimic the signal of a transiting planet,” said Dr. Emily Gilbert, a NASA Sagan Fellow at the University of Chicago who studies exoplanet validation techniques but was not involved in this particular study. “Machine learning approaches are becoming increasingly valuable for helping us sort through these massive datasets systematically.”
TESS monitors hundreds of thousands of stars simultaneously, searching for periodic dimming that could indicate planets passing in front of their host stars. Since beginning operations in 2018, the mission has identified thousands of planetary candidates, but confirming these requires careful analysis to rule out false positives.
The validation process typically involves examining additional observational data, including follow-up observations from ground-based telescopes, analysis of stellar properties, and statistical modeling of potential contaminating sources. Machine learning algorithms can help streamline this process by identifying patterns in the data that distinguish genuine planetary transits from imposters.
According to the study, the AI-assisted approach successfully validated planets ranging from small rocky worlds to gas giants. Among the confirmed planets are several in multi-planet systems, where multiple worlds orbit the same star. These systems are particularly valuable for understanding planetary formation and evolution.
The research also identified several planets in unusual orbital configurations, including some with very short orbital periods. Ultra-short-period planets, which complete their orbits in less than a day, are relatively rare and provide insights into planetary migration and atmospheric evolution under extreme conditions.
Statistical analysis of the validated planets contributes to ongoing efforts to understand planetary population statistics. Studies of exoplanet occurrence rates help astronomers understand how common different types of planetary systems are throughout the galaxy.
Previous research using data from NASA’s Kepler mission found that roughly 20-25% of Sun-like stars host at least one planet smaller than Neptune with an orbital period shorter than 100 days. TESS observations, which survey a much larger portion of the sky than Kepler, are providing additional data to refine these statistical measurements.
The validation work is part of a broader trend toward automated analysis techniques in astronomy. As space-based surveys generate increasingly large datasets, traditional manual analysis methods become impractical.
“We’re in an era where surveys are finding more planet candidates than we can feasibly validate using traditional methods,” said Dr. Jessie Christiansen, project scientist for the NASA Exoplanet Archive at Caltech. “Machine learning tools are becoming essential for processing these large datasets efficiently.”
The Vera C. Rubin Observatory, currently under construction in Chile, will generate approximately 20 terabytes of astronomical data nightly when it begins operations. Similarly large datasets are expected from future space missions, including the European Space Agency’s PLATO mission, planned for launch in the 2020s.
These upcoming surveys will require sophisticated automated analysis pipelines to identify and validate astronomical discoveries, from exoplanets to supernovae to asteroids. Machine learning approaches developed for current missions like TESS provide important groundwork for handling these future data volumes.
The newly validated planets join more than 5,000 confirmed exoplanets cataloged in NASA’s Exoplanet Archive. Since the first detection of planets orbiting Sun-like stars in the 1990s, the field has evolved from discovering individual worlds to conducting statistical surveys of planetary populations.
TESS is expected to continue operations through at least 2025, with the possibility of extended missions depending on spacecraft health and funding availability. The mission has already surveyed roughly 75% of the sky and is beginning its extended mission phase, which will provide additional observations of previously monitored regions.
These repeat observations will enable detection of planets with longer orbital periods and improve the precision of measurements for previously identified candidates. Combined with continuing advances in machine learning techniques, this additional data should yield further discoveries in the coming years.
The research demonstrates the growing importance of interdisciplinary collaboration between astronomy and computer science as astronomical datasets continue to expand in size and complexity.