Create stable R working environment

contains.. 1) setting cran mirror prior to installing R packages 2) coding to sort out categorical & numeric variables 3) boxcox transformation that gives best log-likelihood result 4) fitting normal distribution to y random variable

1. Discussion about 5 function for describing survival model 2. IFR, IFRA

### objective -> quick overview of the data 1. NA check 2. Correlation with response variable -> only numerical variables 3. Correlation within control variables -> check out for multicollinearity

1. Setup A. How to Replace NA -> EDA; you don't have to B. Duplicated Rows -> Eliminate 2. Y variables A. Siginificant correlations ? -> Not Really B. Multi-Collinearity -> Eliminate cast_total_facebook_likes, num_users_review 3. X variables -> Partition in to Categories A. Old Movies B. Independent Films -> Low budget C. Facebook Likes -> if you insists on using, Director_facebook_likes is significant to slight degree; the rest? no

1. Bagging 2. Random Forest

Use Package "randomForest", "tree" 1. Bagging - randomForest( ..., mtry=number of IV, importance=TRUE) 2. Random Forest - mtry, ntree 3. Boosting - distribution="gaussian", n.trees(B), interaction.depth, shrinkage(lambda)

Decision Tree A. Classification Tree B. Regression Tree

Conducting regularization(shrinkage method) w/ Boston Data -> How does ridge, lasso improve a prediction model? => Smaller test MSE A. Compare between Best Subset Selection and Shrinkage method 1) test MSE 2) Variance B. Compare between ridge regression and lasso 1) coefficients 2) variance -> which method was better ?

1. Ridge Regression a) Construct Ridge regression model - Difference with least squares method b) Selecting the best lambda(CV) 2. The Lasso a) Construct Lasso model - Difference with least squares method / ridge regression b) Selecting the best lambda(CV)

EDA Chapter 6_ Probability Plot Examine if sample distribution is from any theoretical distribution -->Compare between sample quantiles and theoretical quantiles

Generating Dependent random variables with copula method 1) Generate Standard normal random variables as intermediary random variables 2) Transform standard random variables using Gaussian Copula 3) Apply resulted values to inversed normal cdf 4) Apply resulted values to inversed exponential cdf

1. Construct Logistic Regression Model 2. Calculate odds and logits for each X 3. k-Fold Cross Validation

Generating Continuous Random variables

1. Symmetrizing Re-expression 2. Variance Stablizing Re-expression

1. Subset Selection A. Best Subset Selection B. Forward Selection and Backward Selection * Results from forward selection and backward selection is almost similar, where as best subset selection yields pretty different results. 2. Cross Validation A. LOOCV B. k-Fold Cross Validation

[Cross Validation Techniques] 1. Leave-One-Out Cross Validation * use cv.glm() * Cross Validation technique with square values of predictor 2. k-Fold Cross Validation [Subset Selection Models] # use regsubsets() from package 'leaps' 1. Best Subset Selection 2. Forward and Backward Stepwise Selection * use argument 'method="forward" or "backward"' in function 'regsubsets()'

Conducting Logistic Regression Analysis with Smarket data A. Construct Logistic Regression model with function glm, argument (family=binomial) B. Function 'predict()'(argument 'type="response"') C. Separating Test data

[Table of Contents] A. Setup B. Construct Regression Model 1) Eliminate Insignificant Predictors 2) Interaction Terms 3) Non-linear Terms C. Diagnosis 1) VIF 2) Plotting - Residual, Leverage

Regression Analysis with Boston Data provided by package MASS 1) Constructing Regression Model a) Add Interaction Terms b) Add non-linear Terms 2) Plotting Regression Model a) Residual - Leverage Plot