Supervised Classification | Google Earth Engine | Google for Developers

Abstraction: Google Earth Engine supervised classification workflow using CART, RandomForest, and SVM

Key points:

• Earth Engine Classifier package supports CART, RandomForest, NaiveBayes, and SVM (via libsvm); training data must be a FeatureCollection with consecutive integer class labels starting at 0
• image.sampleRegions() extracts predictor values from Landsat bands at training point/polygon locations
• Six output modes available: CLASSIFICATION, REGRESSION, PROBABILITY, MULTIPROBABILITY, RAW, RAW_REGRESSION via setOutputMode()
• Accuracy assessment uses ConfusionMatrix; training accuracy overestimates true accuracy — validation split essential
• Spatial autocorrelation can inflate accuracy; spatial join filter removes training samples within 1000m of validation points
• Expensive classifiers can be exported as assets via Export.classifier.toAsset() and reloaded to avoid retraining

Connections: Google · Supervised Learning · Classification · Random Forest · Support Vector Machines

Source: https://developers.google.com/earth-engine/guides/classification