Jane's Garden @janexia215-blog - Tumblr Blog

Bye (at Chicago O'Hare International Airport) https://www.instagram.com/p/Bojh2DjlkEzbJtvlIFbJB6dZxe_L7O5Ez0dkuk0/?utm_source=ig_tumblr_share&igshid=1so0ezryuz81e

Machine Learning for Data Analysis - Week 1 - Sample

SAS Code

Results

#machine learnin #decision trees #sas #coursera

Machine Learning for Data Analysis - Week 1 - Study Notes

Outcome: target or response variables

Explanatory variables: predictors, features or inputs

Unsupervised Learning

Pattern detection or describing association among variables without a specific outcome variable.

Supervised Learning

When the goal is prediction of the value of a response variable based on a number of predictors.

No hypothesis testing

We typically do not go into a machine learning application with specific hypothesis in mind.

We use a subset of observations from our dataset, which we call the training set to learn about the data. And then test the statistical model we get from the training data set on a different set of observations which we call the test set.

Accuracy = Test Error Rate

Goal: find a model that minimizes test error rate. That is, a model that accurately reflects true population associations or patterns.

Linear regression

Accuracy = mean squared error (MSE)

Variance = change in parameter estimated across different data sets

Bias = how far off model estimated values are from true values

We want a statistical model, which has both low variance and low bias. The problem is that these two properties are negatively associated.

Logistic Regression is an example of a classification model.

Accuracy = How well a model correctly classified observations

Decision trees

A data mining method that allow us to explore the prescience of potentially complicated interactions within out data by creating segmentations or sub groups.

When a response variable is categorical, the model is called a classification tree.

Decision trees are read from the top down, starting at the root node.

Growing the Tree

Binary splits maximize correct classification.

All possible separations or cut-points are tested and the separation yielding the minimum impurity or error is selected.

Subgroups showing similar outcomes but different explanatory variable constellations, are generated.

Validating the Tree

Cross-validation guards against overfit.

A random subset is tested and only “branches” that improve the classification are retained.

Select sub-tree is the lowest probability of misclassification.

#machine learning #decision trees #coursera

Regression Modeling in Practice - Week 4 - Sample

Logistic Regression for a Binary Response Variable

SAS Code

Results

#regression #sas #coursera

at Battery Park, Maryland

at National September 11 Memorial & Museum

at Trinity Church Cemetery

at Charging Bull, Wall Street,Manhattan

Regression Modeling in Practice - Week 3 - Sample

Multiple Regression and Confidence Intervals

SAS Code

Results

Polynomial Regression

SAS Code

Results

Evaluating Model Fit

SAS Code

Results

#regression #sas #coursera

Regression Modeling in Practice - Week 3 - Study Notes

Sampling Variability

The sample we draw is not likely to be exactly like the population.

Confidence Interval

Which values of the parameter estimates are plausible in the population.

Typically, we look at 95% confidence interval. With 95% certainty, the range of parameter estimate values includes the true population parameter. That is, we are 95% certain that the true population parameter falls somewhere between the lower and upper confidence limits that are estimated based on our sample parameter estimates.

In linear regression, when you have a non-significant P value, the 95% confidence interval for the parameter estimate will include a value of zero, no association.

Centering is especially important when you are testing a polynomial regression model because it makes it considerably easier to interpret the regression coefficients.

Centering significantly reduces the correlation between the linear and quadratic variables in a polynomial regression model.

Modeling more complex curves in the sample data will often improve the fit of the model for that sample. However, it also increases the risk of overfitting.

An over fitted model may fit the sample really well, but may not fit well on another sample drawn from the same population. An over fitted model is biased toward the sample it was developed on and consequently, the conclusion drawn from that model may not be representative of the population. (Bias-Variance Tradeoff)

#regression #coursera