Multi-Class Logistic Regression 11/07/2018 Liyuan Liu Ph.D. - PowerPoint PPT Presentation

Multi-Class Logistic Regression 11/07/2018 Liyuan Liu Ph.D. Students in Analytics and Data Science Kennesaw State University

Multi-class logistic regression 5-cross validation ROC plot

Process Data Preparation Softmax Function Gradient Descent Model Training and testing Add L1 Regularization 3

Data Preparation 1. Separate raw data to X and Y. 2. Add Intercept. 3. Normalized X use min-max method. 4. One Hot EncodedY. 4

Softmax 1/(1+np.exp(-score)) ( np.exp(score) / np.sum(np.exp(score)) 5

Gradient Descent Learning Rate: 0.01 Epoch: 3000 6

Get Prediction Value Rule: Extract the indexhas the highest probability. The argmax() only for compute accuracy 7

ROC Plot Use numpy.ravel to flatten the array.¶ 8

Result-5 Cross Validation 9

Result L1 Regularization-5 Cross Validation Why Regularization? Reduce Over-fitting Problem. 10

Result L1 Regularization-5 Cross Validation 11

THANKS Liyuan Liu: lliyuan@students.kennesaw.edu