Pima Diabetes Code

Idris Rasheed Pima Native American Code.Rmd

---
title: "Pima Indian Women Code"
author: "Idris Rasheed"
date: "August 12, 2017"
output: pdf_document
---

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

```{r, echo=FALSE}
library(corrplot)
library(e1071)
library(RSNNS)
library(caret)
library(AppliedPredictiveModeling)
library(mice)
library(randomForest)
library(VIM)
library(kernlab)
```

Reading Data 
```{r reading}
#Create a vector of the feature names
headers <- c("TotalPregnancies", "PlasmaGlucose", "DiastolicPressure", "FoldThickness",
"Insulin", "BMI", "PedigreeFunction", "Age", "Diagnosis")
#Import data
url <- paste0("http://archive.ics.uci.edu/ml/machine-learning-databases/",
"pima-indians-diabetes/pima-indians-diabetes.data")
diabetes <- read.csv(url(url),header = FALSE, col.names = headers)
```

Exploratory Analysis
```{r str}
str(diabetes)
```

```{r pair}
pairs(diabetes)
```

Loop
```{r change 0}
#Creates loops to make all 0's NA
for (i in 2:6){
  for (n in 1:nrow(diabetes)){
    if (diabetes[n, i] == 0){
      diabetes[n, i] <- NA
    }
  }
}
```

Factor
```{r factor}
diabetes$Diagnosis <- as.factor(ifelse(diabetes$Diagnosis == 0, "NotDiabetic", "Diabetic"))
```

Density
```{r densities}
transparentTheme(trans = .9)
featurePlot(x = diabetes[, 1:8], 
            y = diabetes$Diagnosis,
            plot = "density", 
            ## Pass in options to xyplot() to 
            ## make it prettier
            scales = list(x = list(relation="free"), 
                          y = list(relation="free")), 
            adjust = 1.5, 
            pch = "|", 
            layout = c(4, 1), 
            auto.key = list(columns = 2))
```

Missing Values
```{r}
table(is.na(diabetes))
```


```{r agg}
aggr(diabetes[,2:6], cex.lab=1, cex.axis = .4, numbers = T, gap = 0)
```

```{r scatter missing}
scattmatrixMiss(diabetes)
```

Predicted Mean Matching
```{r temp data}
com.diabetes <- mice(diabetes, m = 3, method = 'pmm', seed = 100)
```

```{r}
densityplot(com.diabetes)
```

```{r form}
diabetes <- complete(com.diabetes)
```

```{r corr}
corrplot(cor(diabetes[,-9]),type = "lower", method = "number")
```

```{r scale}
diabetes[, 1:8] <- scale(diabetes[, 1:8], center = TRUE, scale = TRUE)
```

Cross-Validation
```{r folds}
Folds <- trainControl(method = "repeatedcv",
                      number = 10,
                      repeats = 10,
                      classProbs=TRUE,
                      summaryFunction=twoClassSummary)
```

Training and Testing
```{r training}
sampleSize <- floor(.7 * nrow(diabetes))
set.seed(100)
Ind <- sample(seq_len(nrow(diabetes)), size = sampleSize)

XTrain <- diabetes[Ind, 1:8]
XTest <- diabetes[-Ind, 1:8]

YTrain <- diabetes[Ind, 9]
YTest <- diabetes[-Ind, 9]
```


Linear Classification
```{r linear svm}
#Creates Linear SVM
linear.tune <- expand.grid(C = c(.1, 1, 10))
set.seed(100)
linear.svm <- train(XTrain,
                    YTrain,
                    method = "svmLinear",
                    metric = "ROC",
                    trControl = Folds,
                    tuneLength = 10,
                    tuneGrid = linear.tune)
linear.svm
```

```{r}
#Linear SVM fit
linear.fit <- svm(Diagnosis ~ ., 
                  data=diabetes,
                  kernel="linear", 
                  cost=.1)
summary(linear.fit)
```
The gamma parameter at 0.125 indicates that the variance is high, but the bias is low. 

##Radial Classfication 
```{r radial svm}
radial.svm.expand <- expand.grid(sigma = c(.2, .4, .6, .8),
C = c(.1, 1, 5, 10, 100))
set.seed(100)
radial.svm <- train(XTrain,
                    YTrain,
                    method = "svmRadial",
                    metric = "ROC",
                    trControl = Folds,
                    tuneGrid = radial.svm.expand)
radial.svm
```

```{r radial fit}
radial.fit <- svm(Diagnosis ~ ., 
                  data=diabetes,
                  kernel="radial", 
                  cost=1,   
                  sigma=0.2)
summary(radial.fit)
```

##Random Forest
```{r randomforest}
rf.expand <- expand.grid(mtry = 2:8)
set.seed(100)
randomforest <- train(XTrain,
                      YTrain,
                      method = "rf",
                      metric = "ROC",
                      trControl = Folds,
                      tuneGrid = rf.expand)
randomforest
```

Plots
```{r rf varimpplot}
varImpPlot(randomforest$finalModel, type = 2, main = "Random Forest")
```

```{r linear plot BMI}
plot(linear.fit, diabetes, PlasmaGlucose ~ BMI)
```

```{r linear TP}
plot(linear.fit, diabetes, PlasmaGlucose ~ TotalPregnancies)
```

```{r radial plot bmi}
plot(radial.fit, diabetes, PlasmaGlucose ~ BMI)
```

```{r radial plot TP}
plot(radial.fit, diabetes, PlasmaGlucose ~ TotalPregnancies)
```

##Accuracy 
```{r rf acc}
#RF Train accuracy
rf.predictTR <- predict(randomforest$finalModel, XTrain)
rf.train.accuracy <- mean(rf.predictTR == YTrain)
rf.train.accuracy
```

```{r rf acc test}
#RF Test accuracy
rf.predictTT <- predict(randomforest$finalModel, XTest)
rf.test.accuracy <- mean(rf.predictTT == YTest)
rf.test.accuracy
```

```{r lin acc}
#Linear SVM Train accuracy
linear.svm.predictTR <- predict(linear.svm$finalModel, XTrain)
linear.svm.train.accuracy <- mean(linear.svm.predictTR == YTrain)
linear.svm.train.accuracy
```

```{r lin acc test}
#Linear SVM Train accuracy
linear.svm.predictTT <- predict(linear.svm$finalModel, XTest)
linear.svm.test.accuracy <- mean(linear.svm.predictTT == YTest)
linear.svm.test.accuracy
```

```{r radial acc}
#Radial SVM Train accuracy
radial.svm.predictTR <- predict(radial.svm$finalModel, XTrain)
radial.svm.train.accuracy <- mean(radial.svm.predictTR == YTrain)
radial.svm.train.accuracy
```

```{r rad test}
#Radial SVM test accuracy
radial.svm.predictTT <- predict(radial.svm$finalModel, XTest)
radial.svm.test.accuracy <- mean(radial.svm.predictTT == YTest)
radial.svm.test.accuracy
```