Abstract
Asteroid taxonomies provide a link to surface composition and mineralogy of those objects, although that connection is not fully unique. Currently, one of the most commonly used asteroid taxonomies is that of Bus-DeMeo. The spectral range covering 0.45–2.45 μm is used to assign a taxonomic class in that scheme. Such observations are only available for a few hundreds of asteroids (out of over one million). On the other hand, a growing amount of space and ground-based surveys delivers multi-filter photometry, which is often used in predicting asteroid types. Those surveys are typically dedicated to studying other astronomical objects, and thus not optimized for asteroid taxonomic classifications. The goal of this study was to quantify the importance and performance of different asteroid spectral features, parameterizations, and methods in predicting the asteroid types. Furthermore, we aimed to identify the key spectral features that can be used to optimize future surveys toward asteroid characterization. Those broad surveys typically are restricted to a few bands; therefore, selecting those that best link them to asteroid taxonomy is crucial in light of maximizing the science output for solar system studies. First, we verified that with the increased number of asteroid spectra, the Bus–DeMeo procedure to create taxonomy still produces the same overall scheme. Second, we confirmed that machine learning methods such as naive Bayes, support vector machine (SVM), gradient boosting, and multilayer networks can reproduce that taxonomic classification at a high rate of over 81% balanced accuracy for types and 93% for complexes. We found that multilayer perceptron with three layers of 32 neurons and stochastic gradient descent solver, batch size of 32, and adaptive learning performed the best in the classification task. Furthermore, the top five features (spectral slope and reflectance at 1.05, 0.9, 0.65, and 1.1 μm) are enough to obtain a balanced accuracy of 93% for the prediction of complexes and six features (spectral slope and reflectance at 1.4, 1.05, 0.9, 0.95, and 0.65 μm) to obtain 81% balanced accuracy for taxonomic types. Thus, to optimize future surveys toward asteroid classification, we recommend using filters that cover those features.
The Challenge of Asteroid Classification
Asteroid taxonomy is the primary tool scientists use to link the light signatures of distant solar system bodies to their mineral composition. The current gold standard, the Bus-DeMeo taxonomy, relies on full spectral coverage across the visible and near-infrared range (0.45–2.45 μm). However, a massive data bottleneck exists: while over a million asteroids have been discovered, high-quality full spectra are available for only a few hundred.
Most asteroid data comes from multi-filter photometry—taking snapshots through specific color filters. Unfortunately, these filters are often designed for galactic or stellar surveys rather than planetary science. This study addresses the need for "minimum viable data" by identifying the specific wavelengths that provide the most diagnostic information for classification, offering a blueprint for optimizing future deep-space surveys.
Methodology and Data Processing
The researchers treated asteroid classification as a supervised machine learning problem, utilizing a combined dataset of 504 asteroids with high-quality spectra from the original Bus-DeMeo (2009) dataset and the MITHNEOS survey (Binzel et al., 2019). The study focused on 12 taxonomic types with sufficient representation (at least 10 objects per class).
The data pipeline involved sophisticated feature engineering. Linear spline fitting was used to smooth noisy spectral data, which was then sampled at 41 discrete points. A critical step was calculating the "spectral slope" (the overall tilt of the spectrum) and subsequently "de-sloping" the data. This process highlights subtle absorption features that are often obscured by the dominant slope, making them more accessible to machine learning algorithms.
Machine Learning and Feature Selection
The study compared five distinct models: Multinomial Logistic Regression, Naive Bayes, Support Vector Machines (SVM), Gradient Boosting (XGBoost), and a Multilayer Perceptron (MLP). To determine feature importance, the team employed Sequential Forward Selection (SFS). This iterative process starts with an empty set and adds the single wavelength that most improves accuracy until performance plateaus.
Dimensionality reduction techniques like Principal Component Analysis (PCA) were also evaluated. As shown in the PCA visualization, different asteroid complexes (such as the stony S-types and carbonaceous C-types) occupy distinct regions of the feature space, though significant overlap remains between individual types, necessitating robust non-linear classifiers like the MLP.
Key Results: The Power of Few Features
The Multilayer Perceptron (MLP) emerged as the superior model, achieving a balanced accuracy of 81% for predicting 12 specific taxonomic types and 93% for predicting the four major asteroid complexes. Several key insights were derived from the feature selection process:
- The 1-μm "Sweet Spot": The most informative region for classification is the 1-μm absorption band, associated with olivine and pyroxene. The top-ranked features cluster tightly between 0.9 and 1.1 μm.
- Slope is King: The spectral slope was consistently identified as the most important individual feature for distinguishing broad complexes.
- The "Magic Six": The authors demonstrated that just six features—the spectral slope plus reflectances at 1.4, 1.05, 0.9, 0.95, and 0.65 μm—are sufficient to reach peak classification performance.
Surprisingly, the 2-μm region, which human experts often use to distinguish types, ranked lower in automated selection. This suggests that the information in the 2-μm band is largely redundant if high-quality data from the 1-μm region and the slope are already available.
Conclusion and Future Directions
This research confirms that the Bus-DeMeo taxonomic scheme is robust and reproducible through automated methods. By identifying a minimal set of diagnostic wavelengths, the study provides a clear recommendation for the design of future astronomical surveys. Implementing filters at these specific wavelengths in upcoming missions, such as the Vera C. Rubin Observatory or the Euclid mission, could exponentially increase our ability to characterize the solar system's asteroid population.
Future work will focus on incorporating albedo (reflectivity) to resolve ambiguities in the X-complex and expanding the models to include rare asteroid types as more data becomes available.
Summary updated