Data Mining for Business Analytics: Concepts, Techniques, and Applications
One book, placed in its field — what Galit Shmueli, Peter C. Bruce, Inbal Yahav, Nitin R. Patel, Kenneth C. Lichtendahl Jr.'s book gets right, where it goes further than the rest, and what the rest of the field adds.
Mike West, Editor · Leading With Data · peopleanalyst.com/guides/profile-data_mining_for_business_analytics_concepts
Arrow keys or swipe to move · N for notes · F for full screen
Data Mining for Business Analytics: Concepts, Techniques, and Applications
One book, placed in its field — what Galit Shmueli, Peter C. Bruce, Inbal Yahav, Nitin R. Patel, Kenneth C. Lichtendahl Jr.'s book gets right, where it goes further than the rest, and what the rest of the field adds.
Mike West, Editor · Leading With Data · peopleanalyst.com/guides/profile-data_mining_for_business_analytics_concepts
The problem
Predictive performance erodes when it is left to instinct instead of method.
You were taught the moves piecemeal, never the whole model.
Data Mining for Business Analytics: Concepts, Techniques, and Applications · Mike West, Editor · Leading With Data
The model
- 01Quality of Problem/Purpose Definition
- 02Data Preparation and Cleaning Quality
- 03Dimension Reduction Effort
- 04Use of Data Partitioning and Cross-Validation
- 05Appropriateness of Method Selection
- 06Model Complexity
- 07Overfitting (Failure to Generalize)
- 08Model Generalization
Data Mining for Business Analytics: Concepts, Techniques, and Applications · Mike West, Editor · Leading With Data
Step 1 of 5 · Quality of Problem/Purpose Definition
Frame the deliverable as a decision that changes, not a variable that gets predicted.
Misconception
Practitioners believe the problem is defined once they know which target variable to predict.
Reality
Naming a target variable is a modeling decision, not a business definition; the real specification is who will use the score, for what decision, at what threshold, and what happens if they act on it.
Data Mining for Business Analytics: Concepts, Techniques, and Applications · Mike West, Editor · Leading With Data
Step 2 of 5 · Data Preparation and Cleaning Quality
Explore distributions before you clean; a summary statistic hides the skew that will break k-NN.
Misconception
Cleaning is a one-time chore you finish before the interesting modeling begins.
Reality
Preparation typically consumes the majority of a project and is iterative — exploring the data reshapes the problem, and modeling failures send you back to fix encodings and outliers you missed.
Data Mining for Business Analytics: Concepts, Techniques, and Applications · Mike West, Editor · Leading With Data
Step 3 of 5 · Dimension Reduction Effort
Fewer, well-chosen predictors often generalize better than a large kitchen-sink set.
Misconception
More variables give the model more to learn from, so reduction is a reluctant compromise for speed.
Reality
Beyond a point, added dimensions dilute the density of your data so that neighbors become meaningless and spurious correlations look real; reduction often improves accuracy, not just runtime.
Data Mining for Business Analytics: Concepts, Techniques, and Applications · Mike West, Editor · Leading With Data
Step 4 of 5 · Use of Data Partitioning and Cross-Validation
Partition first, transform second — anything fit on the full data contaminates the holdout.
Misconception
The validation set is just for reporting final performance once the model is chosen.
Reality
Every time you use the validation set to pick a model or tune a parameter, you leak it into training; that is why a third, untouched test set (or nested cross-validation) is needed for a truly honest estimate.
Data Mining for Business Analytics: Concepts, Techniques, and Applications · Mike West, Editor · Leading With Data
Step 5 of 5 · Appropriateness of Method Selection
No algorithm dominates across problems; the fit depends on your data's structure and your explanation needs.
Misconception
There is a single best algorithm, and the job is to find and apply it.
Reality
Methods coexist because each encodes different assumptions about the data — linear boundaries, local neighborhoods, feature independence, interactions — and the best choice depends on which assumptions your data actually satisfies.
Data Mining for Business Analytics: Concepts, Techniques, and Applications · Mike West, Editor · Leading With Data
Data Mining for Business Analytics: Concepts, Techniques, and Applications · Mike West, Editor · Leading With Data
The cheat sheet
- Quality of Problem/Purpose DefinitionFrame the deliverable as a decision that changes, not a variable that gets predicted.
- Data Preparation and Cleaning QualityExplore distributions before you clean; a summary statistic hides the skew that will break k-NN.
- Dimension Reduction EffortFewer, well-chosen predictors often generalize better than a large kitchen-sink set.
- Use of Data Partitioning and Cross-ValidationPartition first, transform second — anything fit on the full data contaminates the holdout.
- Appropriateness of Method SelectionNo algorithm dominates across problems; the fit depends on your data's structure and your explanation needs.
- Model ComplexityThe right complexity is where validation error bottoms out, not where training error does.
- Overfitting (Failure to Generalize)Low training error with high holdout error is the definition of a model that will disappoint in deployment.
- Model GeneralizationGeneralization is measured on unseen data drawn from the deployment population, not just any holdout.
Data Mining for Business Analytics: Concepts, Techniques, and Applications · Mike West, Editor · Leading With Data
Sources
- Data Mining for Business Analytics: Concepts, Techniques, and Applications — Galit Shmueli, Peter C. Bruce, Inbal Yahav, Nitin R. Patel, Kenneth C. Lichtendahl Jr.
Data Mining for Business Analytics: Concepts, Techniques, and Applications · Mike West, Editor · Leading With Data