Spatial and Predictive Analytics
Spatial tools operate on spatial objects carried in a field of type SpatialObj. Key tools: Create Points (from lat/long), Spatial Match (point-in-polygon, distance), Trade Area (drive-time or radius), Buffer, Poly-Build, and Find Nearest. A common trap: Spatial Match requires both inputs to contain spatial objects; a text lat/long field must first pass through Create Points. Another: coordinate system mismatches (WGS84 vs projected) produce wrong distances.
Predictive tools include Linear Regression, Logistic Regression, Decision Tree, Forest Model, Score, and Cross Validation. Understand the split between estimation (training) and scoring (applying the model). The Score tool consumes an output model object; the Cross Validation tool partitions data and reports average error. The Field Summary and Data Investigation tools help profile before modeling.
Traps: regression requires numeric predictors (categorical fields must be encoded), and missing values cause rows to drop silently. Logistic regression predicts a binary target, not a continuous one. Overfitting shows as strong training accuracy but weak cross-validation accuracy.
Rule of thumb: always hold out data and compare training vs validation error before trusting a model.
Worked example: predict churn (Yes/No) from tenure and spend. Use Logistic Regression with Churn as target, tenure/spend as predictors, then Score new customers. If accuracy is 95% train but 60% validation, the model is overfit - reduce predictors or add regularization.