Random Forest Classifier
Definition
A supervised machine-learning ensemble model that builds multiple decision trees from randomly sampled subsets of training data and features, then aggregates their outputs by majority vote. Used in OSAC-validated ML tools for ASTM E1618 class assignment in fire debris GC-MS chromatograms.
- Type
- Supervised ensemble machine-learning model
- Core method
- Aggregates predictions from multiple decision trees via majority vote
- Forensic applications
- Accelerant classification in fire debris, geological sample sourcing
Common questions
What makes a random forest classifier different from a single decision tree?+
A random forest builds many decision trees instead of just one, each trained on different random subsets of the data. It then combines all their predictions by majority vote, which produces more reliable results and reduces the risk of overfitting to quirks in the training data.
Where is random forest used in forensic science?+
It's applied to fire debris analysis for classifying accelerant compounds in GC-MS chromatograms and to geological samples for pinpointing regional source areas. OSAC-validated ML tools use this method for standardized class assignments in accelerant detection.
How does a random forest handle uncertainty?+
When applied to geological data, the classifier outputs not just a predicted source region but also a posterior probability, giving investigators confidence in how certain the match is. This probabilistic output is more useful than a hard yes-or-no answer in cases where the sample is ambiguous.
Related terms
- Cal Fire (California Department of Forestry and Fire Protection)
- California's state fire agency, which has integrated drone aerial survey and multispectral imaging into major wildfire investigations since 2018. Operational use in...
- Detection Limit
- The lowest concentration an analytical method can reliably distinguish from background noise. Data below the detection limit are censored, and combining datasets...
- ENFSI EWG AI Working Group
- The European Network of Forensic Science Institutes Explosives Working Group's artificial intelligence working group, active from 2021. Produced a 2023 position paper...
- FOREGS
- Forum of European Geological Surveys. Its geochemical baseline surveyed approximately 900 stations across Europe for more than 50 elements in stream water,...
- G-BASE
- The BGS Geochemical Baseline Survey of the Environment. Soil and stream-sediment samples collected across Britain at roughly one per 2 km2, analysed...
- Geochemical Province
- A geographically coherent region where soils share a distinctive elemental signature attributable to a common bedrock, mineralisation, or depositional history. Knowing which...
- Handheld Raman
- A portable spectrometer using a near-infrared laser (typically 785 nm or 1064 nm) to measure inelastic light scattering from molecular bonds. Generates...
- LIBS (Laser-Induced Breakdown Spectroscopy)
- A spectroscopic technique that uses a high-power pulsed laser to ablate and ionise a small amount of sample material, generating a plasma...
- Multispectral Imaging
- Photography or remote sensing in multiple wavelength bands beyond the visible spectrum, including near-infrared and short-wave infrared. Drone-mounted multispectral cameras are used...
- OSAC F&E Subcommittee
- The Organisation of Scientific Area Committees Fire Investigation and Explosives Subcommittee, operating under NIST. Produces method standards, validation guidance, and proficiency criteria...
- Photogrammetric 3D Scene Model
- A three-dimensional spatial reconstruction of a scene produced by photogrammetric processing of overlapping photographs, typically captured by drone. Treated by courts as...
- Structure-from-Motion (SfM)
- A photogrammetric technique in which overlapping images from a drone or camera, processed by software, generate a three-dimensional point cloud and textured...
Explained in these topics
- Emerging Methods: Handheld Raman, LIBS, CT Imaging, MLA supervised machine-learning ensemble model that builds multiple decision trees from randomly sampled subsets of training data and features, then aggregates t...
- Forensic Geology Reference DatabasesAn ensemble machine-learning method that builds many decision trees from bootstrapped subsets of the training data and combines their outputs. Applied to geoch...