Decision Tree - Theory (Math), Application & Modeling with R

Explore model design principles using real-world data, linking historical window data to upgrade responses to build predictive models with predictor and dependent variables.

Explore model design precautions, including selecting variables from historical and effect windows that are forecast time–appropriate; exclude unavailable, cherry-picked, or extremity-affected data and ensure a stable performance window.

Explains how to define model design outcomes in credit modeling, covering observation windows, exclusions, bad definitions, and in-time versus out-of-time validation with a 70/30 development split.

Design the performance window long enough for performance to mature, with at least 200 good and 200 bad accounts and a rolling window for equal timeframes.

Audit quality by ensuring 0 and 1 outcomes, handle missing values, treat indeterminants, convert numerics to categories, apply capping and flooring, and group values to avoid cherry-picking for decision trees.

Demonstrate a decision tree development in R Studio, explore a dataset, install the R package and R Studio, and build your first decision tree in R with detailed code steps.

Explore the dt_02_02_credit_1.xls data for decision tree development with 1000 observations and variables like chk_acct, duration, history, and new_car to predict the binary response 0 or 1 using R.

Learn how to access course data files, such as excel or sas code, by using the view resource and the download area to reveal and download linked materials.

Install the R statistical package and the R Studio GUI from cran.r-project.org and rstudio.com, then explore the four-panel interface for decision tree modeling.

Learn how to develop a decision tree in R using rpart and partykit, import data with read.csv, and configure model parameters like min split, min bucket, and max depth.

Explore building a second decision tree in R for classification using rpart, converting 0/1 responses to Y/N, and interpreting leaf nodes, splits, and misclassification loss.

The course adds practical usage of classification and regression trees, with demos, assignments, and a model solution between modules to enhance learning and job readiness.

Use classification trees in R to convert wine quality data into a binary yes/no outcome, and explore splits like alcohol under 10.85 to guide leaves.

This assignment tests your ability to interpret classification tree outputs using blood donation data, guiding you to convert 1/0 to yes/no and reproduce results with set.seed(100) and a party object.

Build and interpret a classification tree in R with rpart and partykit: read CSV, recode Y/N, fit and plot, and interpret root and leaf splits and KS metrics.

Explore the decision tree algorithm: how it develops from categorical outcomes using the gini method, trains classification trees in R, and extends to regression trees and the chi square method.

Explore how decision trees choose numeric variable splits and where to cut, using percentile-based cuts, evaluating 20 numeric variables and 1,980 possible splits, and identifying the best split by formula.

Explain the Gini index as a node impurity measure in decision trees using the formula 1 minus the sum of squared class proportions, with examples.

Compute the Gini index of a split to measure impurity in a decision tree, using the weighted Gini of left and right; aim for 0 as best split.

discover how a decision tree selects the best variable and cut by minimizing the gini index of a split to reduce impurity, including numeric cuts and categorical splits.

Develop a practical decision tree in data mining by setting max depth to 4, minimum leaf observations, and up to 16 nodes, then validate with development and validation data.

Compare in-time and out-of-time validation for decision tree models using historical data, and analyze population and bad-rate shifts across nodes to assess model robustness.

Apply a decision tree model to unseen data by following training rules in a data mining scenario, using chk_acct, duration, and sav_acct to assign a response rate.

Explore auto pruning of decision trees, trimming weak branches using validation data, preventing overfitting, and identifying the optimal leaf-node level with CART, Gini, and k-fold validation in R.

Explore k-fold cross-validation, where data are randomly divided into k parts, train on k-1 parts, and validate on the remaining part, repeated across all folds (k=5 or 10).

Develop a regression tree in R by using R-squared to determine splits, implement CART with ANOVA, and interpret a height and weight example and the resulting node structure.

Interpret a regression tree output by examining height and weight splits, terminal nodes, and box plot visuals, then assess model fit with deviance, y values, and R-square.

Develop a regression tree in R using CART and ANOVA to model mileage from price, country, reliability, and type, with categorical handling and pruning insights.

Develop and interpret a regression tree in R using partykit, handle missing values, clean data, convert variables, and examine R-squared and node insights.

Identify the average mileage for displacements between 190.5 and 284.5 and interpret the regression tree's root split by weight at 2217, including medians and range in box plots.

Develop and interpret regression trees using eight input variables to predict heating load and cooling load for twelve building shapes, focusing on patterns and model outputs.

Construct two regression trees for heating and cooling load from eight input variables in r, using rpart and partykit, and interpret model outputs, splits, and r-squared values.

Compare linear regression and regression trees, noting that regression trees handle categorical inputs, missing data, and outliers more robustly, while linear regression may offer a better fit when applicable.

Explore the CHAID algorithm, a chi-square automatic interaction detector, which uses chi-square on contingency tables to build decision trees and assess relationships via observed versus expected counts and p-values.

Compare CHAID and CART: CHAID uses p-values to select splits and allows multiway nodes with a forward stopping rule, while CART minimizes impurity with binary splits and pruning.

Explore the chi square test of independence and chi square statistic for how gender, region, and education affect success rates. Learn how observed versus expected differences drive interpretation and dependence.

Explore the chi square formula, including observed, expected, and the numerator and denominator. Learn why squaring magnifies divergence and makes values positive for meaningful comparison.

Learn how degree of freedom in contingency tables affects chi-square significance, using the (r-1)*(c-1) formula for 2x2 and larger tables.

Explore the chi square distribution, derive that chi square statistics are nonnegative, understand the mean and variance equals 2 df, and interpret critical values and significance across degrees of freedom.

Explore the decision tree algorithm family by analyzing ID3 using entropy, including entropy of a node and splits, with Excel-assisted calculations, then introduce random forest modeling in R.

Explore entropy of a node as a measure of impurity, using base-2 logarithms and relative frequencies to quantify purity between defaulter and non-defaulter classes.

Master how the ID3 method uses entropy to evaluate splits across numeric and categorical variables, selecting the best variable and cut value, maximizing impurity reduction to grow the tree.

Explore how the random forest method builds many random decision trees from data and variable subsets, forming a forest that improves classification through majority voting.

Install random forest package in R, load it with library, and build a forest model using a dataset and target tilde, then inspect fit for confusion matrix and variable importance.

Explore ensemble learning, including bagging and boosting, where parallel models vote for the final prediction and sequential models adjust weights for each new learner, highlighting k-fold validation.

Complete the decision tree course using R, apply the concepts confidently, share feedback to help learners, and explore courses in logistic regression, cluster analysis, and data analysis using Excel.