Predictive and Statistical Analytics
This domain covers the Predictive tool category and Summarize-based statistics. Know the difference between supervised (classification/regression) and unsupervised (clustering) tools. Linear Regression outputs a model and a report; Logistic Regression predicts a binary target; Decision Tree and Forest Model handle classification/regression; K-Centroids Cluster Analysis and K-Nearest Neighbors are unsupervised/similarity tools. Score applies a saved model to new data. Summarize supports Count, Count Non Null, Sum, Average, Min, Max, and Percentile actions; Crosstab and Frequency Table summarize categories.
Critical preprocessing traps: categorical fields must be encoded (dummy variables) before regression; missing values must be handled (Impute) or the tool drops rows; target leakage inflates accuracy. Train/test splits (via Create Samples) prevent overfitting claims. The Model Comparison tool compares candidate models. Reading output matters: coefficients, p-values, R-squared for regression; confusion matrix, accuracy, precision/recall for classification.
Worked example: predicting churn (binary). Steps: clean nulls, convert categorical Plan to dummy variables with Create Indicators, split 70/30 with Create Samples, fit Logistic Regression, then Score the holdout and inspect the confusion matrix. Rule of thumb: Never report training accuracy as model performance - always score a holdout set.