Machine Learning Using SPSS Modeler Training Course

Participants can import data and configure fields for reliable analysis in Modeler.

• Import flat files, spreadsheets, databases and IBM SPSS Statistics data
• Database source nodes configured with credentials, queries and field metadata
• Type node measurement levels set for categorical and continuous fields
• Record identifiers separated from predictors before analytical sampling begins
• Metadata reviewed for storage, precision, formats and value labels
• Data provenance captured from source extraction through stream output

Participants can diagnose data-quality issues and define appropriate remediation actions.

• Data Audit node profiles distributions, missingness, extremes and field quality
• Frequency tables expose invalid categories, sparse levels and coding errors
• Histograms and quantiles identify skewness, outliers and implausible values
• Missing-value patterns distinguished between structural, random and systematic absence
• Duplicate records detected using identifiers, matching keys and aggregation checks
• Audit findings converted into documented remediation priorities and assumptions

Participants can prepare consistent development datasets and create reproducible data partitions.

• Filter, Select and Sample nodes create defined analytical populations
• Derive node expressions recode values, parse strings and calculate ratios
• Missing values imputed using median, mode, constants or predictive methods
• Outliers treated through capping, transformation, exclusion or indicator flags
• Categorical predictors regrouped to retain signal and adequate record counts
• Partition node creates training, testing and validation samples reproducibly

Participants can create and assess predictors that represent business behaviour and timing.

• Date functions derive tenure, recency, seasonality and elapsed-time predictors
• Aggregation nodes create customer, account and transaction-level behavioural features
• Binning converts continuous variables into interpretable risk or response bands
• Interaction terms represent conditional effects between products, channels and demographics
• Feature selection combines domain hypotheses, correlation checks and model importance
• Leakage checks prevent post-outcome information entering predictor fields

Participants can construct development samples suitable for rare-event prediction problems.

• Stratified sampling preserves rare events across development and validation partitions
• Class imbalance diagnosed using event rates, confusion matrices and prevalence
• Balance node applies oversampling, undersampling and adjusted class weights
• Cost matrices express unequal consequences of false approvals and false declines
• Prior probabilities adjusted when development samples differ from production populations
• Sampling decisions recorded for reproducible model comparison and approval

Participants can develop classifiers and convert predicted probabilities into operational decisions.

• Binary, multiclass and ordinal targets matched to suitable modeling methods
• Classifier node compares multiple algorithms under common partition settings
• Probabilities, predicted classes and confidence values interpreted for operational decisions
• Confusion matrices quantify true positives, false positives and classification errors
• Thresholds selected from business costs, capacity constraints and risk appetite
• Lift charts assess ranking performance for response and fraud models

Participants can build interpretable tree models and translate splits into business rules.

• CHAID trees split categorical targets using statistically significant predictor interactions
• C&R Tree models create binary splits using impurity reduction rules
• C5.0 produces rule sets, boosting options and predictor importance
• QUEST trees reduce selection bias among predictors with many categories
• Tree depth, minimum records and pruning control model complexity
• Terminal nodes translated into operational rules and customer segments

Participants can compare interpretable and nonlinear methods for classification problems.

• Logistic regression estimates event probabilities and interpretable predictor coefficients
• Bayesian networks represent conditional dependencies among predictors and target classes
• Support vector machines separate classes using maximum-margin decision boundaries
• Neural networks model nonlinear patterns through hidden-layer transformations
• Auto Classifier compares eligible methods and ranks candidate models
• Model choices balanced between accuracy, transparency, speed and maintenance

Participants can build and validate models that predict numeric business outcomes.

• Linear regression predicts continuous outcomes using estimated coefficient relationships
• Generalized linear models match distributions and link functions to targets
• Regression assumptions checked through residuals, influence and multicollinearity diagnostics
• R-squared, RMSE and MAE compared across holdout samples
• Continuous predictions converted into planning ranges and decision thresholds
• Auto Numeric selects competing models using consistent partition performance

Participants can select models using statistical evidence and business decision criteria.

• Evaluation node compares models using shared targets, partitions and metrics
• ROC curves and AUC measure discrimination independent of selected thresholds
• Gain and lift charts quantify campaign efficiency at ranked depths
• Calibration plots compare predicted probabilities with observed event rates
• Profit matrices test expected value under alternative decision policies
• Champion-challenger selection documents performance, interpretability and implementation constraints

Participants can assess ensemble models and justify their use against simpler alternatives.

• Bagging reduces variance by aggregating models trained on resampled records
• Boosting sequentially focuses modeling effort on previously misclassified observations
• Random forests combine randomized trees for stable nonlinear classification
• Ensemble outputs assessed against single models using holdout evidence
• Predictor importance compared cautiously across tree and ensemble methods
• Complexity justified against response time, auditability and maintenance requirements

Participants can create stable segments and translate profiles into differentiated actions.

• K-Means clustering partitions records by distance to centroid profiles
• TwoStep clustering selects cluster numbers using statistical fit criteria
• Kohonen networks map similar records into neighbourhood-based segment structures
• Input scaling prevents high-range variables dominating distance calculations
• Cluster profiles interpreted through demographics, behaviours and outcome measures
• Segment stability tested across samples and alternative cluster specifications

Participants can identify useful product, event and process relationships from transactional data.

• Apriori association rules identify product combinations from transaction records
• Support, confidence and lift distinguish frequent rules from useful rules
• Sequence rules detect ordered events within customer or process histories
• Association inputs prepared through transaction identifiers, items and event timestamps
• Rule pruning removes redundant patterns and operationally unusable combinations
• Cross-sell actions prioritised using rule strength, volume and margin

Participants can transform unstructured text into features for analysis and prediction.

• Text Analytics extracts concepts, categories and sentiments from unstructured responses
• Term extraction settings control linguistic resources, synonyms and irrelevant terms
• Concept patterns convert complaints and survey comments into usable predictors
• Text-derived fields combined with structured data for predictive modeling
• Category quality reviewed against representative documents and subject-matter labels
• Privacy risks assessed before processing free-text personal information

Participants can score new data and produce action lists from approved models.

• Model nugget scores new records using stored transformations and parameters
• Scoring branches retain identical preparation steps used during model development
• Generated fields named for predicted values, probabilities and confidence
• PMML exports support portable deployment where compatible scoring engines exist
• Batch scoring outputs reconciled with input counts and exception records
• Decision rules route scored cases to campaigns, reviews or interventions

Participants can document model controls, limitations and accountability requirements.

• Model documentation records purpose, data sources, assumptions and limitations
• Data minimisation restricts fields to justified analytical and operational purposes
• Protected characteristics assessed for proxy variables and disparate outcomes
• Access controls separate development data, credentials and approved production streams
• Versioned approvals establish accountable ownership for models and scoring changes
• Human review points defined for high-impact or low-confidence decisions

Participants can define controls for detecting declining model performance after deployment.

• Population stability index detects shifts between development and live populations
• Performance monitoring tracks discrimination, calibration and realised business outcomes
• Drift thresholds trigger investigation, recalibration, retraining or model retirement
• Champion-challenger tests compare replacement models against current production performance
• Monitoring schedules align refresh frequency with data volatility and decision risk
• Incident logs capture failures, overrides, complaints and corrective actions

Participants can present model findings and define practical implementation requirements.

• Model narratives connect analytical evidence to decisions, controls and expected value
• Visualisations present segment profiles, lift, risk thresholds and uncertainty
• Executive summaries distinguish predictive performance from realised operational benefit
• Implementation plans specify owners, data feeds, actions and acceptance criteria
• Decision logs record approved thresholds, exceptions and escalation procedures
• Model handover packs support operational teams after analyst transition

Participants can build, review and present a complete IBM SPSS Modeler machine learning solution for a defined business decision.

• Project brief defines decision, target, users, constraints and success measures
• End-to-end stream imports, audits, prepares and partitions supplied business data
• Candidate models compared using appropriate metrics, costs and validation evidence
• Selected model scored into an action list with documented thresholds
• Governance pack records assumptions, limitations, monitoring metrics and ownership
• Presentation defends recommendations to stakeholders using model results and business implications