Loading Libraries and Data

library(tidyverse)
library(caTools)
library(MASS)
library(dplyr)
happydata<-read.csv("C:/Users/power/BST260/Project/BST260_ClassProject/5years.csv",header=T)

set.seed(123)
dataset <- dplyr::select(happydata,HScore,Economy,Family,Life.expectancy,Freedom,Generosity,Trust)

split = sample.split(dataset$HScore, SplitRatio = 3/4)
training_set = subset(dataset, split == TRUE)
test_set = subset(dataset, split == FALSE)

After these thorough analyses of the previous session. We would ask how we can predict the happiness session? These session, three different machine learning stratigies are used to predict the happiness scores. The section aims to find which strategy would be the best one.

library(tidyverse)
library(dslabs)
library(broom)
regressor_lm = lm(formula = HScore ~ ., data = training_set)

Test-data

y_pred_lm = predict(regressor_lm, newdata = test_set)
Pred_Actual_lm <- as.data.frame(cbind(Prediction = y_pred_lm, Actual = test_set$HScore))

Plot

gg.lm <- ggplot(Pred_Actual_lm, aes(Actual, Prediction )) +
  geom_point(color="blue",size = 3, alpha = 0.5) + theme_bw() + geom_abline(col="red", lwd=1, lty=2) +stat_smooth(method = "lm", col = "blue") +
  labs(title = "Multiple Linear Regression", x = "Actual happiness score",
       y = "Predicted happiness score") +
  theme(plot.title = element_text(family = "Helvetica", face = "bold", size = (15)), 
        axis.title = element_text(family = "Helvetica", size = (10)))
  
gg.lm

Neural net

library(neuralnet)

nn <- neuralnet(HScore ~ Economy + Family + Life.expectancy + Freedom + Generosity + Trust,
                data=training_set,hidden=1,linear.output=TRUE)
predicted.nn.values <- compute(nn,test_set[,2:7])

Pred_Actual_nn <- as.data.frame(cbind(Prediction = predicted.nn.values$net.result, Actual = test_set$HScore))

gg.nn <- ggplot(Pred_Actual_nn, aes(Actual, V1 )) +
    geom_point(color="blue",size = 3, alpha = 0.5) + theme_bw() + geom_abline(col="red", lwd=1, lty=2) +stat_smooth(method = "lm", col = "blue") +
  labs(title = "Neural Net", x = "Actual happiness score",
       y = "Predicted happiness score") +
  theme(plot.title = element_text(family = "Helvetica", face = "bold", size = (15)), 
      axis.title = element_text(family = "Helvetica", size = (10)))
gg.nn

Decision Tree

# Fitting Decision Tree Regression to the dataset
library(rpart)
regressor_dt = rpart(formula = HScore ~ .,
                  data = dataset,
                  control = rpart.control(minsplit = 10))
# Predicting a new result with Decision Tree Regression
y_pred_dt = predict(regressor_dt, newdata = test_set)

Pred_Actual_dt <- as.data.frame(cbind(Prediction = y_pred_dt, Actual = test_set$HScore))


gg.dt <- ggplot(Pred_Actual_dt, aes(Actual, Prediction )) +
    geom_point(color="blue",size = 3, alpha = 0.5) + theme_bw() + geom_abline(col="red", lwd=1, lty=2) +stat_smooth(method = "lm", col = "blue") +
  labs(title = "Decision Tree Regression", x = "Actual happiness score",
       y = "Predicted happiness score") +
  theme(plot.title = element_text(family = "Helvetica", face = "bold", size = (15)), 
        axis.title = element_text(family = "Helvetica", size = (10)))
gg.dt

Based on the Mean Sqaure Error and the five year data, neural net has the lowest value of MSE, which indicate that neural net could be the best one to predict happiness score. However, multiple linear regression has the very close value of MSE. In addition, the result of multiple linear regression is much more easily to be interpreted. Thus, both neural net and multiple linear regression are recommended to predict the happines score.

MSE.lm <- sum((test_set$HScore - y_pred_lm)^2)/nrow(test_set)
MSE.nn <- sum((test_set$HScore - predicted.nn.values$net.result)^2)/nrow(test_set)
MSE.dtr<-sum((test_set$HScore - y_pred_dt)^2)/nrow(test_set)

print(paste("Mean Squared Error (Multiple Linear Regression):", MSE.lm))
[1] "Mean Squared Error (Multiple Linear Regression): 0.308477276098853"
print(paste("Mean Squared Error (Neural Net):", MSE.nn))
[1] "Mean Squared Error (Neural Net): 0.302149143404976"
print(paste("Mean Squared Error (Decision Tree Regression):", MSE.dtr))
[1] "Mean Squared Error (Decision Tree Regression): 0.385762435452679"