IntroToDataMining : Key Components and process

IntroToDataMining : Key Components and process

• 1.
  Data Mining Overview
• 2.
  Core Ideas inData Mining  Classification - Classifying the outcome of categorical variable based on historical data  Prediction – prediction of a given output variable (target variable) based on other variables or features  Association Rules & Recommenders  Data & Dimension Reduction  Data Exploration  Visualization
• 3.
  Supervised Learning  Predictivemodeling describes supervised learning - the training data we feed the algorithm includes the target variable. We are “supervising” the training process  Training data, where target value is known  Score to data where value is not known  Methods: Classification and Prediction
• 4.
  Supervised: Classification  Theobject of classification is to predict a categorical variable  Examples: Purchase/no purchase, fraud/no fraud, creditworthy/not creditworthy…  Target variable is often binary (yes/no)
• 5.
  Supervised: Prediction  Goal:Predict numerical target (outcome) variable  Examples: sales, revenue, performance  As in classification:  Each row is a case (customer, tax return, applicant)  Each column is a variable  Taken together, classification and prediction constitute “predictive analytics”
• 6.
  Unsupervised Learning  Performsan analysis without a target variable  Goal: Segment data into meaningful segments; detect patterns  There is no target (outcome) variable to predict or classify  Example: Clustering – segments the observations into to similar groups based on the variables
• 7.
  Unsupervised: Association Rules Goal: Produce rules that define “what goes with what”  Example: “If X was purchased, Y was also purchased”  Rows are transactions  Used in recommender systems – “Our records show you bought X, you may also like Y”  Also called “affinity analysis”
• 8.
  Unsupervised: Data Reduction Distillation of complex/large data into simpler/smaller data  Reducing the number of variables/columns (e.g., principal components)  Reducing the number of records/rows (e.g., clustering)
• 9.
  Unsupervised: Data Visualization Graphs and plots of data  Histograms, boxplots, bar charts, scatterplots  Especially useful to examine relationships between pairs of variables
• 10.
  Data Exploration  Datasets are typically large, complex & messy  Need to review the data to help refine the task  Use techniques of Reduction and Visualization
• 11.
  The Process ofData Mining
• 12.
  Steps in DataMining 1. Define/understand purpose 2. Obtain data (may involve random sampling) 3. Explore, clean, pre-process data 4. Reduce the data; if supervised DM, partition it 5. Specify task (classification, clustering, etc.) 6. Choose the techniques (regression, neural networks, etc.) 7. Iterative implementation and “tuning” 8. Assess results – compare models 9. Deploy best model
• 13.
  Test Partition  Whena model is developed on training data, it can overfit the training data (hence need to assess on validation)  Assessing multiple models on same validation data can overfit validation data  Some methods use the validation data to choose a parameter. This too can lead to overfitting the validation data  Solution: final selected model is applied to a test partition to give unbiased estimate of its performance on new data
• 14.
  Data Splitting  Dataare split into training and test sets.  Training data is used to train the model  Testing data is used to estimate an unbiased assessment of the model’s performance #Random splitting of iris data into 70% train and 30%test datasets ind <- sample(2, nrow(data), replace=TRUE, prob=c(0.7, 0.3)) trainData <- data[ind==1,] testData <- data[ind==2,]
• 15.
  Simple Random Sampling All observations have an equal chance of selection # Using base R set.seed(123) # for reproducibility index_1 <- sample(1:nrow(ames), round(nrow(ames) * 0.7)) train_1 <- ames[index_1, ] test_1 <- ames[-index_1, ]
• 16.
  Rare Event Oversampling Often the event of interest is rare  Examples: response to mailing, fraud in taxes, …  Sampling may yield too few “interesting” cases to effectively train a model  A popular solution: oversample the rare cases to obtain a more balanced training set  Later, need to adjust results for the oversampling
• 17.
  Stratified Sampling  Commonin classification problems where the response variable may be severely imbalanced  Example: 90% of the observations have a “yes” response and 10% have a “no” response.  Solution: segment into quartiles and then random sample from each quartile set.seed(123) split_strat <- initial_split(attrition, prop = 0.7, strata = "Attrition") train_strat <- training(split_strat) test_strat <- testing(split_strat)