FeatureEng_GermanCredit_Template_All.Rmd

---
title: "Feature Engineering German Credit - Steve's Attempt"
output: 
  html_document:
      toc: yes
      toc_float: yes
      code_folding: hide
---

```{r setup, include=FALSE}
knitr::opts_chunk$set(echo = TRUE)
```


```{r}
library(dplyr)
library(caret)
library(rpart)
library(rpart.plot)
```

# Load Data

```{r}
data(GermanCredit, package = "caret")
df = GermanCredit
df$Class = as.character(df$Class)
df$Class[df$Class == "Bad"] = "NotGood" # Rename, just for personal preference.
df$Class = as.factor(df$Class)
str(df)
head(df)
table(df$Class)

formula = Class ~ .
positive = "Good"
```


# Feature Engineering

```{r}
# Hint: use the preProcess() and predict() fkunctions
p1 <- preProcess(df)
df = predict(p1, df)
```

# Feature Selection

```{r}
dim(df)
set.seed(10)

# The following snippet will perform feature selection using caret's SBF = Selection By Filtering
filterCtrl <- sbfControl(functions = rfSBF, number=1, verbose=T)
r <- sbf(formula, data = df, sbfControl = filterCtrl)
r
df = cbind(df[,predictors(r)], Class=df$Class)
dim(df)
```

# Splitting the Data

```{r}
set.seed(123) # Set the seed to make it reproducible

train.index <- createDataPartition(df$Class, p = .8, list = FALSE)
train <- df[ train.index,]
test  <- df[-train.index,]

# Double check that the stratefied sampling worked
table(df$Class)/nrow(df)
table(train$Class)/nrow(train)
table(test$Class)/nrow(test)

actual = test$Class
```

# Trying it all: KNN

```{r}


ctrl <- trainControl(method = "repeatedcv", 
                     number = 10, repeats = 5, 
                      classProbs = TRUE, returnResamp = "all", sampling="down")

kknn_grid <- expand.grid(.kmax = c(3, 5, 10), 
                    .distance=c(1, 2), 
                    .kernel=c("rectangular", "gaussian", "biweight"))

kknn_fit <- train(formula, 
                  data = train, 
                  method = "kknn", 
                  metric="Kappa", 
                  preProc=c('nzv', 'center', 'scale'),
                  trControl=ctrl, tuneGrid = kknn_grid)

summary(kknn_fit)
pred = predict(kknn_fit, test)
caret::confusionMatrix(data=pred, reference=actual, positive=positive, dnn=c("Predicted", "Actual"))
```

# Trying it all: RF

```{r, warning=FALSE}
ctrl <- trainControl(method = "repeatedcv", 
                     number = 10, repeats = 5, 
                      classProbs = TRUE, returnResamp = "all", sampling="down")

rf_grid <- expand.grid(.mtry = c(50, 100, 500))

rf_fit <- train(formula, data = train, "parRF",
                  preProc=c('nzv', 'center', 'scale'),
                  trControl = ctrl, tuneGrid = rf_grid, 
                  metric="Kappa", allowParallel = TRUE)

summary(rf_fit)
pred = predict(rf_fit, test)
caret::confusionMatrix(data=pred, reference=actual, positive=positive, dnn=c("Predicted", "Actual"))
```