PredictWages.Rmd

---
title: "Analyze & Predict Hourly Wages"
author: "Sarvani Vadali"
date: "April 24, 2016"
output: html_document
---

1. loading required packages
---
*Uncomment the follwing install commands to install packages if needed*
```{r echo=TRUE, message=FALSE, warning=FALSE}
#install.packages("klaR")
#install.packages("e1071")
#install.packages("dplyr")
#install.packages("data.table")
#install.packages("e1071")
#install.packages("caret")
#install.packages("klaR")
#install.packages("randomForest")

require(klaR)
require(e1071)
require(dplyr)
require(ggplot2)
require(data.table)
require(car)
require(randomForest)

library(plyr)
library(dplyr)
library(data.table)
library(e1071)
library(caret)
library(ggplot2)
```

# 2. Load dataset as a data frame
*The following script assumes that **SLID-Ontario.txt** file is present in the current working directory*

```{r}
# Cleanup workspace.
rm(list=ls())

# Load data file as a data frame.
wages.data <- read.table("SLID-Ontario.txt", header = TRUE)
wages.data <- na.omit(wages.data)
```
#### 2.1 View structure of the data

```{r}
summary(wages.data)

pairs(wages.data)

# Plot and visually inspect if gender plays a role in wages.
plot(wages.data$sex, wages.data$compositeHourlyWages)
```


*It seems like gender plays a role in hourly wages. This can be further analyzed using more advanced techniques.*

# 3. Analysis
## 3.1 Regression model

#### 3.1.1 Age vs Hourly wage
*Determine how hourly wage changes over age*

```{r}

# Build a simple regression model to analyze how age determines hourly wage.
wageAndAge.lmfit	= lm(wages.data$compositeHourlyWages ~ wages.data$age)
summary(wageAndAge.lmfit)
plot(fitted(wageAndAge.lmfit), residuals(wageAndAge.lmfit),
  xlab = "Fitted Values", ylab = "Residuals")
  abline(h=0, lty=2)
  lines(smooth.spline(fitted(wageAndAge.lmfit), residuals(wageAndAge.lmfit)))
residualPlots(wageAndAge.lmfit)
```

The values are reasonably densed around the center line and the model can be used to perform intial analysis.

**It seems like hourly wages increase over age in general but tend to decline over late years**

#### 3.1.2 Gender vs Hourly wage

*Determine if gender plays a role in hourly wages*

```{r}
# Build a simple regression model to analyze how gender determines hourly wage.
genderAndEducation.lmfit	= lm(wages.data$compositeHourlyWages ~ wages.data$sex, data = wages.data)
summary(genderAndEducation.lmfit)

plot(wages.data$compositeHourlyWages~wages.data$sex, data=wages.data, main="Gender vs Hourly wage", xlab = "Gender", ylab = "Hourly wage")
abline(genderAndEducation.lmfit, col="red")
```

**It seems gender plays a significant role in hourly wages. Men tend to getting paid more then women.**

#### 3.1.2 Years of education vs Hourly wage

```{r}
# Build a simple regression model to analyze how years of education determines hourly wage.

wageAndEducation.lmfit	= lm(wages.data$compositeHourlyWages ~ wages.data$yearsEducation, data = wages.data)
summary(wageAndEducation.lmfit)

plot(wages.data$compositeHourlyWages~wages.data$yearsEducation, data=wages.data, main="Education vs Hourly wage", xlab = "Education", ylab = "Hourly wage")
abline(wageAndEducation.lmfit, col="red")

```

**It seems that as number of years of education increases pay in general increases**

## 3.2 Naive-Bayes 
*Perform further analysis using Naive-Bayes algorithm*

#### 3.2.1 Manipulate data

*Add range of wages to existing data*

```{r}
min.wage = floor(min(wages.data$compositeHourlyWages))
max.wage = ceiling(max(wages.data$compositeHourlyWages))
num.of.levels = 10 # Add ten wage ranges.
level.range = ceiling((max.wage - min.wage) / num.of.levels)

# Add new levels to dataset.
wages.data$wageRange <- ""
x <- min.wage
y <- level.range

repeat {
  wages.data[wages.data$compositeHourlyWages >= x & wages.data$compositeHourlyWages < y ,]$wageRange = paste(x,"To",y)
  x <- y
  y <- x + level.range
  
  if(y > max.wage){
    break
  }
}

wages.data.new <- wages.data
wages.data.new$compositeHourlyWages <- NULL # Remove composite data.

```

#### 3.2.1 Naive-Bayes classification model

*Create trainig and testing data*

```{r}
set.seed(1234)
index <- createDataPartition(wages.data.new$wageRange, p = .8, list = FALSE)
wages.training.data <- wages.data.new[index, ]
wages.testing.data <- wages.data.new[-index, ]
```

*Create model*

```{r}
nb.model <- naiveBayes(as.factor(wageRange)~., data = wages.training.data)

# exploring the nb object:
names(nb.model)

nb.pred <- predict(nb.model, newdata=wages.testing.data, laplace=3)
```

*Predictions for gender vs wages*

```{r}
table(nb.pred, wages.testing.data$sex)
```

*Predictions for age vs wages*

```{r}
table(nb.pred, wages.testing.data$age)
```

*Predictions for years of education vs wages*

```{r}
table(nb.pred, wages.testing.data$yearsEducation)
```

**Naive Bayes also confirms that gender plays a role in determining hourly wages**

#### 3.2.2 Support Vector Machine model

*Build model*

```{r echo=TRUE, message=FALSE, warning=FALSE}
set.seed(5678)

fit.Control <- trainControl(method = "repeatedcv",
  number = 10,
  repeats = 10)

svm.model <- train(wageRange ~ ., data = wages.training.data,
                 method = "svmLinear",
                 trControl = fit.Control,
                 verbose = FALSE)
svm.model

svm.pred <- predict(svm.model, newdata=wages.testing.data, laplace=3)
```

*Predictions for gender vs wages*

```{r}
table(svm.pred, wages.testing.data$sex)
```

*Predictions for age vs wages*

```{r}
table(svm.pred, wages.testing.data$age)
```

*Predictions for years of education vs wages*

```{r}
table(svm.pred, wages.testing.data$yearsEducation)
```