Udemy
    •  
    •  
    •  
    •  
    •  
    •  
    •  
    •  
Subscribe
Turn what you know into an opportunity and reach millions around the world.
Learn More
Your cart is empty.
Keep shopping
Mastering and Tuning Decision Trees
Rating: 3.6 out of 5(39 ratings)
383 students

Mastering and Tuning Decision Trees

IBM SPSS Modeler Seminar Series
Created bySandy Midili
Last updated 7/2014
English
English [Auto],

What you'll learn

  • Understand the theory behind classification trees
  • Differentiate between classification tree algorithms
  • Know the assumptions of classification trees
  • Learn the advantage and disadvantages of the different algorithms
  • Interpret the results

Course content

2 sections • 24 lectures • 3h 3m total length

  • Characteristics of Tree Models2:51

    Describe supervised segmentation in decision trees using a historical data set of correct and incorrect matches. Contrast with unsupervised segmentation; discuss non-tree rule-generation technique and deployable rules.

  • Supervised Segmentation in Modeler2:59

    Analyze how supervised segmentation uses targets to differentiate churn among loyalty program customers, and compare decision trees with rule-based models using the auto classifier and expert tab.

  • Trees and Rules2:30

    Explore how a rule like store cards equals yes and spend ratio above 0.4 informs purchase, and why decision lists differ from trees by not being mutually exclusive or exhaustive.

  • Defining Terms2:49

    Review root and leaf (terminal) nodes; sum leaf samples to the tree’s grand total. Contrast decision lists with trees; unclassified remainder lowers accuracy, while trees are exhaustive and mutually exclusive.

  • Additional Uses of Trees2:38

    Discover how decision trees support data reduction, reveal interactions in exploration, enable data augmentation, and guide merging of nominal categories to improve model accuracy.

  • A First Look at the Data Set2:50

    Explore the Titanic data set, its variables such as passenger id, cabin, age, and family counts, and learn measurement levels and train and test data sets for decision tree modeling.

  • CHAID4:57

    Explore CHAID, the chi-square automatic interaction detection method, an early data mining technique, and learn how it handles scale variables, missing data as a separate category, and settings.

  • How does the CHAID algorithm work36:58

    Discover how CHAID uses chi-square tests to split data, adjust with Bonferroni, and handle nominal, ordinal, and scale variables in exhaustive versus standard CHAID.

  • Using Autoclassifer to run multiple CHAID settings4:41

    Use the auto classifier to run multiple CHAID settings and compare Pearson and likelihood ratio approaches, noting chi-square assumptions and trade-offs between exhaustive and faster searches on test data.

  • How does the C&RT algorithm work?12:31

    Discover how CART uses binary splits with purity and balance, highlights the Gini coefficient, and handles nominal, ordinal, and scale variables, including pruning.

  • Surrogates for Missing Data12:32

    Explore surrogates for missing data in decision trees, using the five variables most correlated with age as proxies to decide branches, and tune their number to balance accuracy and speed.

  • C&RT's Expert Settings19:14

    Discover how pruning, stopping rules, and minimum change in impurity keep cart trees from overfitting by balancing growth with pruning through cost complexity analysis.

  • CHAID Expert Settings9:40

    Explore CHAID expert settings to control pruning, re-splitting, and thresholds, compare CHAID's tree changes with CART, and assess overfitting versus prediction-focused analysis.

  • Audience Question: How do you compare the results of two models?2:14

    Learn how to compare two models by chaining their predictions, observing how partitions affect outcomes and confidence, and auto-compare without reconfiguring the trade model.

  • Should We Adjust the Models12:25

    Explore how partition choices, sample size, and pruning affect decision tree stability and accuracy, and compare bagging, boosting, and data prep to build robust ensembles.

  • What is Bagging and Boosting?7:50

    Explore how bagging creates multiple models, why data prep matters before bagging, and how boosting and ensemble methods influence accuracy and stability in decision trees.

Requirements

  • Knowledge or experience with IBM SPSS Modeler or completion of an introductory level data mining course and on the job data mining experience.

Description

IBM SPSS Modeler is a data mining workbench that allows you to build predictive models quickly and intuitively without programming. Analysts typically use SPSS Modeler to analyze data by mining historical data and then deploying models to generate predictions for recent (or even real-time) data.

Overview: Mastering and Tuning Decision Trees is a series of self-paced videos that discusses the decision tree methods (CHAID, C5.0, CRT, and QUEST) available in IBM SPSS Modeler. These techniques produces a rule based predictive model for an outcome variable based on the values of the predictor variables. Students will gain an understanding of the situations in which one would this technique, its assumptions, how to do the analysis automatically as well as interactively, and how to interpret the results. Particular emphasis is made on contrasting CHAID and C&RT in detail. Tuning – the adjusting of parameters to optimize performance – is demonstrated using both CHAID and C&RT.

Who this course is for:

  • Anyone that has experience with IBM SPSS Modeler or has completed an introductory level data mining course and would like to learn about decision tree models.

Report abuse