Session 124TS, A Practical Guide to Machine Learning for Actuaries. Presenters: Dave M. Liner, FSA, MAAA, CERA

Transcription

1 Session 124TS, A Practical Guide to Machine Learning for Actuaries Presenters: Dave M. Liner, FSA, MAAA, CERA SOA Antitrust Disclaimer SOA Presentation Disclaimer

2 A practical guide to machine learning for actuaries Dave Liner 27 JUNE 2018

3

4

5 CareerCast top job rankings by year Actuary Rank

6 CareerCast top job rankings by year Actuary Data Scientist Statistician Rank

7

8

9

10 9

11

12

13 What I learned about machine learning 1. Machine learning drives disruption and innovation in many sectors globally 2. Many actuaries already do most of the machine learning process 3. Most popular machine learning methods are based on concepts that actuaries already understand 4. There is a staggering amount of free resources to develop machine learning skills 12

14 Actuaries already do most of the machine learning process 1. Get data 2. Prepare data 3. Build model 4. Use model to gain insight 5. Tell others about model results 13

15 We will review five machine learning methods today k-nearest neighbors k-means clustering Decision trees Linear regression Logistic regression Neural networks 14

16 Illustrative dataset 15

17 X1 = Petal length X2 = Petal width Y = Species 16

18 It is easier to illustrate methods using 2-dimensional data Petal Width Setosa Versicolor Virginica Petal Length 17

19 k-nearest neighbors 18

20 What is the species of a new sample based on Petal width and length using knn? Petal Width Setosa Versicolor Virginica Petal Length 19

21 What is the species of a new sample based on petal width and length using knn? Petal Width Setosa Versicolor Virginica Petal Length 20

22 Four lines of Python code gets same result as Excel knn file import sklearn.datasets as ds, sklearn.model_selection as ms, sklearn.preprocessing as pp, sklearn.neighbors, sklearn.metrics X_train,X_test,y_train,y_test = ms.train_test_split(pp.scale(ds.load_iris().data),ds.load_iris().target,test_size=30,random_state=0) y_pred_knn = sklearn.neighbors.kneighborsclassifier(n_neighbors=3).fit(x_train,y_train).predict(x_test) print(sklearn.metrics.confusion_matrix(y_test, y_pred_knn)) 21

23 k-means clustering 22

24 What is the species of a new sample based on Petal width and length using clustering? Petal Width Unsupervised learning does not require labels (Y) Petal Length 23

25 Step 1: select random centroids (k=3 in this case) Petal Width Petal Length 24

26 Step 2: move centroids based on data Petal Width Petal Length 25

27 Step 3: keep moving centroids until no points are reassigned Petal Width Petal Length 26

28 Step 4: assign each point to a centroid cluster Petal Width Petal Length 27

29 Step 4: assign each point to a centroid cluster Petal Width Setosa Versicolor Virginica Petal Length 28

30 Step 5: use clusters to assign new data points Petal Width Setosa Versicolor Virginica Petal Length 29

31 Step 5: use clusters to assign new data points Petal Width Setosa Versicolor Virginica Petal Length 30

32 Step 5: use clusters to assign new data points Petal Width Setosa Versicolor Virginica Petal Length 31

33 Decision trees 32

34 33

35 Decision trees provide another method for classification Root Node Decision Node Setosa Versicolor Virginica Leaf Nodes 34

36 Decision trees provide another method for classification Petal Width Setosa Versicolor Virginica Petal Length 35

37 Decision trees provide another method for classification Petal Width Setosa Versicolor Virginica Petal Length 36

38 Decision trees provide another method for classification Petal Width Setosa Versicolor Virginica Petal Length 37

39 Decision trees provide another method for classification Petal Width Setosa Versicolor Virginica Petal Length 38

40 Decision trees provide another method for classification Petal Width Setosa Versicolor Virginica Petal Length 39

41 Decision trees provide another method for classification Petal Width Setosa Versicolor Virginica Petal Length 40

42 Logistic regression 41

43 Linear regression uses line of best fit to make numerical predictions Petal Width Petal Length 42

44 Logistic regression is a classification algorithm Yes Is it Virginica? Setosa Versicolor Virginica No Petal Width 43

45 Is it virginica? Yes Virginica? 1.0 Petal Width Setosa Versicolor Virginica No Petal Width Petal Length 44

46 Is it virginica? Yes Virginica? 1.0 Petal Width Setosa Versicolor Virginica No Petal Width Petal Length 45

47 Neural networks 46

48 Neural can model non-linear situations Input Hidden Output Petal Length Sepal Length Petal Width Sepal Width Activation Functions Setosa Versicolor Virginica 47

49 Neural Networks Input Layer Hidden Layer Output Layer 48

50 Neural Networks Input Layer Hidden Layer Output Layer 49

51 Neural Networks Input Layer Hidden Layer Output Layer 50

52 Neural Networks Input Layer Hidden Layer Output Layer 51

53 Summary of methods Unsupervised Deep Regression k-nearest Neighbors k-means Clustering Decision Trees Linear Regression Logistic Regression Neural Networks 52

54 Next steps 53

55

56 Possible machine learning applications in healthcare include Predicting non-adherent drug event before it happens Predict opioid drug abuse before it happens Estimate member persistency based on member and dependent characteristics Project medical costs using personal and clinical data Develop clinical best-practices by linking clinical and financial data 55