HEART DISEASE PREDICTION

ARVIND PAWAR

AZAR NAJAFLI

DHRUVIN KOTHARI

NORTHEASTERN UNIVERSITY

ALY6015 INTERMEDIATE ANALYTICS

INSTRUCTOR NAME- TENGLONG LI

DATE- 03/28/2019
 INTRODUCTION
Cardiology is one of the most important and yet very difficult field of health care. Heart
diseases can prove lethal if not detected in early stages. Every year around 610000 people die from
heart diseases in the United States of America, which is 1 in every 4 deaths, according to the data
provided by CDC (CDC, 2017). The same report says that the disease is main cause of death for
both men and women, however more than half of similar diseases have been observed in men in
2009. Every year more than 700000 Americans have heart attacks in which more than 500000
have it for the first time and more than 200000 have already have attack before. Because of these
statistics, every year researches are done by different people including doctors all around the world
and data analysts/scientists. The role of data science in those researches is that doc
With all the information we had above, we decided to choose this topic as part of our final
project in which our main objective is to construct a predictive model that will help us to determine
the disease beforehand any anomalies in patient’s health.
Our team used the dataset called heart.csv received from Kaggle.com which is an open
source web site for various real-life datasets. Our dataset contains 14 variables (columns) and 304
observations (rows). We decided to define the variable called “target” as dependent one which
means, our model will predict that variable using others (independent). As the variable “target” is
binomial which is either yes or no (1 – if the patient has heart disease or 0 – the patient does not
have a heart disease), the best model to use should be logistic regression.
RStudio is used for all our analysis and visualizations.
We installed all the required packages.
install.packages("DataExplorer") library(ggplot2
install.packages("Hmisc")
install.packages("data.table")
install.packages("caret")
install.packages("extrafont")
install.packages("ggthemes")
install.packages("caret",
repos = "http://cran.r-project.org",
dependencies = c("Depends", "Imports", "Suggests"))
> library(caret)
> library(data.table)
> library(Hmisc)
> library(DataExplorer)
> library(ggplot2)
> library(carData)
> library(car)
> library(dplyr)
> library(lattice)
> library(tidyr)
> library(caret)
> library(MASS)
> library(broom)
> library(ROCR)
> library(corrplot)
> setwd("C:/Users/Arvind/Desktop/Projects/Heart disease/")
> heart_data= read.csv("heart.csv")
> head(heart_data)
ï..age sex cp trestbps chol fbs restecg thalach exang oldpeak slope ca thal target
1 63 1 3 145 233 1 0 150 0 2.3 0 0 1 1
2 37 1 2 130 250 0 1 187 0 3.5 0 0 2 1
3 41 0 1 130 204 0 0 172 0 1.4 2 0 2 1
4 56 1 1 120 236 0 1 178 0 0.8 2 0 2 1
5 57 0 0 120 354 0 1 163 1 0.6 2 0 2 1
6 57 1 0 140 192 0 1 148 0 0.4 1 0 1 1
> names(heart_data)
[1] "ï..age" "sex" "cp" "trestbps" "chol" "fbs" "restecg" "thalach" "exang"
[10] "oldpeak" "slope" "ca" "thal" "target"
> colnames(heart_data)[colnames(heart_data)=="ï..age"]<- "age"
Renamed the age column name
Some of the variables of our data has different categories for example chest pain has type
0,1,2,3, ECG level 0,1,3 and so on. To determine the effect of each variable and better results
we decided to perform factorization.
> factor_heart_data<- copy(heart_data)
> colnames(factor_heart_data)[colnames(factor_heart_data)=="ï..age"]<- "age"
> head(factor_heart_data)
age sex cp trestbps chol fbs restecg thalach exang oldpeak slope ca thal target
1 63 1 3 145 233 1 0 150 0 2.3 0 0 1 1
2 37 1 2 130 250 0 1 187 0 3.5 0 0 2 1
3 41 0 1 130 204 0 0 172 0 1.4 2 0 2 1
4 56 1 1 120 236 0 1 178 0 0.8 2 0 2 1
5 57 0 0 120 354 0 1 163 1 0.6 2 0 2 1
6 57 1 0 140 192 0 1 148 0 0.4 1 0 1 1
>col_names<-
c("age","sex","chest_pain","rest_bp","chol","fasting_bloodsugar","rest_ecg","max_heartrate",
+ "exercise_angina","ST_depression","slope","n_major_vasel","thal","target")
> names(factor_heart_data)<- col_names
> names(factor_heart_data)
[1] "age" "sex" "chest_pain" "rest_bp" "chol"
[6] "fasting_bloodsugar" "rest_ecg" "max_heartrate" "exercise_angina" "ST_depression"
[11] "slope" "n_major_vasel" "thal" "target"
> factor_heart_data$sex <- as.character(heart_data$sex)
> factor_heart_data$sex <- ifelse(heart_data$sex=="0", 'female', 'male')
> factor_heart_data$chest_pain<-as.factor(heart_data$cp)
> factor_heart_data$fasting_bloodsugar[heart_data$fasting_bloodsugar == 1]= "Diabetic"
> factor_heart_data$fasting_bloodsugar[heart_data$fasting_bloodsugar == 0] = "Normal"
> factor_heart_data$rest_ecg[heart_data$rest_ecg == 0] = "Normal"
> factor_heart_data$rest_ecg[heart_data$rest_ecg == 1] = "Abnormality"
> factor_heart_data$rest_ecg[heart_data$rest_ecg == 2] = "Probable or definite"
> factor_heart_data$exercise_angina[heart_data$exercise_angina == "1"]= "yes"
> factor_heart_data$exercise_angina[heart_data$exercise_angina == "0"] = "no"
> factor_heart_data$slope=as.factor(heart_data$slope)
> factor_heart_data$thal=as.factor(heart_data$thal)
> factor_heart_data$target=as.factor(heart_data$target)
> factor_heart_data$sex=as.factor(heart_data$sex)
> factor_heart_data$fasting_bloodsugar=as.factor(heart_data$fbs)
> factor_heart_data$exercise_angina=as.factor(heart_data$exang)
> #View(factor_heart_data)
> head(factor_heart_data)
age sex chest_pain rest_bp chol fasting_bloodsugar rest_ecg max_heartrate exercise_angina
ST_depression slope
1 63 1 3 145 233 1 0 150 0 2.3 0
2 37 1 2 130 250 0 1 187 0 3.5 0
3 41 0 1 130 204 0 0 172 0 1.4 2
4 56 1 1 120 236 0 1 178 0 0.8 2
5 57 0 0 120 354 0 1 163 1 0.6 2
6 57 1 0 140 192 0 1 148 0 0.4 1
n_major_vasel thal target
1 0 1 1
2 0 2 1
3 0 2 1
4 0 2 1
5 0 2 1
6 0 1 1
Our response variable is “target” that has binary data, 0 and 1. 0 means a person don’t have heart
disease and 1 means a person has heart disease.
> plot_histogram(factor_heart_data)
> plot_density(factor_heart_data)

> plot_correlation(heart_data)
From the above results we can say that Age, Cholesterol, maximum heart rate and blood pressure
follows normal distribution.
We started to plot the correlation matrix to see the relationship between “target” and all
other variables visually. As we had, relatively large number of independent variables, we decided
to focus on 4 (cholesterol, heart rate, chest pain, and rest ECG test (this is the test advised by
doctors if the patient has those three symptoms)) which doctors pay attention more when they try
to determine the illness. Below you can see the correlation matrix:
Surprisingly, we can see from the matrix that, although the doctors have more focus on those 4
variables we mentioned above, cholesterol, heart rate, and rest ECG test do not have strong
correlation with dependent variable.

VISUALIZATION
Age Distribution:
> ggplot(factor_heart_data,aes(age, fill=target)) +
+ geom_histogram(aes(y=..density..),breaks=seq(0, 80, by=1), color="grey17") +
+ geom_density(alpha=.1, fill="black")+
+ facet_wrap(~target, ncol=1,scale="fixed") +
+ scale_fill_manual(values=c("blue","red"))+
+ xlab("Age") +
+ ylab("Density / Count") +
+ ggtitle("Age Distribution")
Output:
Conclusion:
From the above graph we can conclude that, heart disease is uniformly spread out across Age.
Gender count for heart disease:
colnames(heart_data)[colnames(heart_data)=="ï..age"]<-"Age"
head(heart_data)
ggplot(factor_heart_data,aes(target, fill=target)) +
geom_bar(stat = "count") + facet_wrap(sex~.) + scale_fill_manual(values=c("Blue","red")) +
ggtitle("Gender count for Heart Disease")
Output:
Conclusion:
From the above graph we can conclude that, more females have Heart Disease as compared to
male patients.
2) To construct a predictive model, we focused performing analysis on the following parameters.
 Chest Pain (Diabetic Patient): As the chest pain increases there is higher probability
that the patient will have a heart problem.
> ggplot(factor_heart_data,aes(target,fill=target)) +
+ geom_bar(stat = "count") + facet_wrap(chest_pain~.) + ggtitle("Count of Heart Patients having
different types of chest Pains") + theme_bw() +
+ scale_fill_manual(values=c("Blue","red"))+
+ xlab("Target")
Output:

Conclusion: More Heart Disease patients have chest pain type 1 or 2

 Cholesterol Level: As the cholesterol level increases chances of having stroke, diabetes
or high blood pressure also increases which may lead to severe heart problems.
> ggplot(factor_heart_data,aes(factor_heart_data$chol, fill=target)) +
+ geom_histogram(aes(y=..density..),breaks=seq(90, 550, by=25), color="grey17") +
+ geom_density(alpha=.1, fill="black")+
+ facet_wrap(~target, ncol=1,scale="fixed") +
+ theme_economist() +
+ scale_fill_manual(values=c("blue","red")) +
+ xlab("Cholestoral levels") +
+ ylab("Density / Count") +
+ ggtitle("Cholestrol level test")
Output:

Conclusion: More Heart Disease patients seem to have between 200 and 250 mg/dl.
 Heart Rate Monitoring (Blood Pressure): For the patients having blood high blood
pressure are often considered as higher probability of having heart problems.
> ggplot(factor_heart_data,aes(factor_heart_data$max_heartrate, fill=target)) +
+ geom_histogram(aes(y=..density..),breaks=seq(70, 205, by=10), color="grey17") +
+ geom_density(alpha=.1, fill="black")+
+ facet_wrap(~target, ncol=1,scale="fixed") +
+ theme_economist() +
+ scale_fill_manual(values=c("blue","red"))+
+ xlab("Maximum Heart Rate Achieved") +
+ ylab("Density / Count") +
+ ggtitle("Max Heart Rate Histogram")
Output:

Conclusion: Heart Disease patients have higher maximum heart rate than healthy patients
 Rest ECG Tests: ECG test are carried out to monitor heart conditions.
> ggplot(factor_heart_data,aes(target,fill=target)) +
+ geom_bar(stat = "count") + facet_wrap(rest_ecg~.)+
+ ggtitle("Patients with different types of ECG test")+
+ scale_fill_manual(values=c("Blue","red"))
Output:

Conclusion: From the above graph we can observe that Patients with Rest ECG 1 have more
probability of having Heart Diseases.
 PREDICTION
We are going to examine the relationship between one categorical response variable and
various categorical independent variables, so we chose Logistic Regression for prediction.
We divided the dataset into training and testing data in 75% and 25% respectively.
> set.seed(12345)
> train <- floor(0.75*nrow(factor_heart_data))
> train_ind <-sample(seq_len(nrow(factor_heart_data)),size = train)
> trainset <- factor_heart_data[train_ind, ]
> testset <- factor_heart_data[-train_ind, ]
> dim(trainset)
[1] 227 14
> dim(testset)
[1] 76 14
Above are the dimensions of our training and testing dataset.
First, we applied the logistic regression model on all the independent variables using glm()
function.
> logit<-glm(target~., data=trainset, family = binomial)
> summary(logit)
Call:
glm(formula = target ~ ., family = binomial, data = trainset)
Deviance Residuals:
Min 1Q Median 3Q Max
-2.6911 -0.3425 0.1598 0.6103 2.6969
Coefficients:
Estimate Std. Error z value Pr(>|z|)
(Intercept) 1.258e+01 1.455e+03 0.009 0.993106
age 6.877e-03 2.703e-02 0.254 0.799172
sex1 -1.677e+00 5.846e-01 -2.868 0.004133 **
chest_pain1 8.921e-01 5.946e-01 1.500 0.133532
chest_pain2 2.034e+00 5.486e-01 3.708 0.000209 ***
chest_pain3 1.947e+00 7.560e-01 2.575 0.010016 *
rest_bp -1.469e-02 1.315e-02 -1.117 0.263945
chol -5.670e-03 4.236e-03 -1.339 0.180657
fasting_bloodsugar1 -2.108e-01 6.912e-01 -0.305 0.760370
rest_ecg1 4.284e-01 4.259e-01 1.006 0.314494
rest_ecg2 -4.029e-01 2.361e+00 -0.171 0.864495
max_heartrate 2.541e-02 1.220e-02 2.083 0.037226 *
exercise_angina1 -6.633e-01 4.795e-01 -1.383 0.166528
ST_depression -4.378e-01 2.672e-01 -1.638 0.101338
slope1 -1.322e-01 8.605e-01 -0.154 0.877894
slope2 7.741e-01 9.387e-01 0.825 0.409615
n_major_vasel -8.687e-01 2.348e-01 -3.700 0.000216 ***
thal1 -1.124e+01 1.455e+03 -0.008 0.993840
thal2 -1.176e+01 1.455e+03 -0.008 0.993551
thal3 -1.302e+01 1.455e+03 -0.009 0.992861
---
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
(Dispersion parameter for binomial family taken to be 1)
Null deviance: 312.74 on 226 degrees of freedom
Residual deviance: 160.10 on 207 degrees of freedom
AIC: 200.1
Number of Fisher Scoring iterations: 14
From above output, we observed that sex, chest pain, number of major vessels observed,
max heart rate has effect on heart disease.
Then, we decided to create another dataset that has significant variables only.
> cor_data<-trainset[,c(2,3,9,10,12,14)]
> summary(cor_data)
sex chest_pain exercise_angina ST_depression n_major_vasel target
0: 96 0:143 0:204 Min. :0.00 Min. :0.0000 0:138
1:207 1: 50 1: 99 1st Qu.:0.00 1st Qu.:0.0000 1:165
2: 87 Median :0.80 Median :0.0000
3: 23 Mean :1.04 Mean :0.7294
3rd Qu.:1.60 3rd Qu.:1.0000
Max. :6.20 Max. :4.0000
However, before applying the model let’s see the variance inflation factor to check whether
there is multicollinearity or not.
> #variance inflation factor
> vif(glm(target ~ ., data=cor_data, family="binomial"))
GVIF Df GVIF^(1/(2*Df))
sex 1.056310 1 1.027770
chest_pain 1.311732 3 1.046263
exercise_angina 1.093148 1 1.045537
ST_depression 1.152195 1 1.073403
n_major_vasel 1.045553 1 1.022523
VIF seems to be low, and there is no multicollinearity in data. So, we can move forward to
apply the logit model on significant variables data.
> logit1<-glm(target~., data=cor_data, family=binomial)
> summary(logit1)
Call:
glm(formula = target ~ ., family = binomial, data = cor_data)
Deviance Residuals:
Min 1Q Median 3Q Max
-2.3277 -0.5202 0.2011 0.5714 2.5038
Coefficients:
Estimate Std. Error z value Pr(>|z|)
(Intercept) 1.9614 0.4348 4.511 6.44e-06 ***
sex1 -1.4117 0.3894 -3.625 0.000289 ***
chest_pain1 1.3498 0.4868 2.773 0.005560 **
chest_pain2 2.0905 0.4192 4.987 6.12e-07 ***
chest_pain3 2.0161 0.6086 3 .313 0.000924 ***
exercise_angina1 -1.2217 0.3721 -3.283 0.001028 **
ST_depression -0.8060 0.1810 -4.454 8.42e-06 ***
n_major_vasel -0.7635 0.1662 -4.595 4.34e-06 ***
---
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
(Dispersion parameter for binomial family taken to be 1)
Null deviance:
417.64 on 302 degrees of freedom
Residual deviance: 238.32 on 295 degrees of freedom
AIC: 254.32
Number of Fisher Scoring iterations: 5
From the above results we can see that all the variables are significant, which is good for
prediction. Let’s visualize the correlation.
> cor_data$chest_pain<-as.factor(cor_data$chest_pain)
> factor_heart_data$chest_pain<-as.factor(heart_data$cp)
> logit1.df<-tidy(logit1)
> library(ggthemes)
> library(extrafont)
> logit1.df %>%
+ mutate(term=reorder(term,estimate)) %>%
+ ggplot(aes(term,estimate, fill=estimate))+
+ geom_bar(stat="identity")+
+ ggtitle("Effect of variables resulting Heart Disease")+
+ scale_fill_gradient(low="blue", high="red")+
+ theme_economist()+
+ geom_hline(yintercept=0)+
+ coord_flip()
Output:

From the above output, we observed that chest pain type 2 has the most impact on heart
disease. Person having any type of chest pain is likely to have heart disease. Increase in
number of major vessels and exercise angina reduces the chances of heart disease.
Our varianvce inflation factor result was good but still to avoid the problem of overfitting
we performed cross validation using trainControl() function.
> fitControl <- trainControl(method = "repeatedcv",
+ number = 10,
+ repeats = 10,
+ classProbs = TRUE,
+ summaryFunction = twoClassSummary)
> trainset$target<-make.names(trainset$target)
> set.seed(142)
> trainset$target<-as.factor(trainset$target)
> generalized_model <- caret::train(target ~ .,
+ data = trainset ,
+ method = "glm",
+ trControl = fitControl,
+ metric="ROC")
There were 20 warnings (use warnings() to see them)
> generalized_model
Generalized Linear Model
227 samples
13 predictor
2 classes: 'X0', 'X1'
No pre-processing
Resampling: Cross-Validated (10 fold, repeated 10 times)
Summary of sample sizes: 205, 204, 205, 205, 205, 203, ...
Resampling results:
ROC Sens Spec
0.8685239 0.7367273 0.8478205
> pred <- predict(generalized_model, testset,type='raw')
> summary(pred)
X0 X1
33 43
Here we passed the test dataset for prediction using predict() function.
> pred
[1] X1 X1 X1 X1 X1 X1 X1 X1 X1 X1 X1 X1 X1 X1 X1 X1 X1 X1 X1 X1 X1 X1 X1 X1 X1
X0 X0 X1 X1 X1 X1 X1 X1 X1 X0 X1
[37] X1 X1 X1 X1 X1 X0 X0 X0 X0 X0 X0 X1 X0 X0 X0 X0 X0 X0 X0 X0 X0 X0 X0 X0 X0
X1 X0 X0 X0 X0 X0 X0 X0 X1 X1 X0
[73] X0 X1 X0 X0
Levels: X0 X1
> testset$target
[1] 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 0 0 0 0 0 0 0 0 0
0000
[55] 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
Levels: 0 1
We can compare the predicted values and observed values.
Here, we created a confusion matrix
> Confusion_Matrix<-table(testset$target, pred)
> Confusion_Matrix
pred
X0 X1
0 30 5
1 3 38
In the end we determined the accuracy of our model.
> accuracy<-sum(diag(Confusion_Matrix))/sum(Confusion_Matrix)
> accuracy
[1] 0.8947368
We got 89.47% accuracy.

CONCLUSION
We analyzed various parameters that risk in causing heart disease, and we created a predictive
model with 89.47% accuracy that can help us determine beforehand whether preventive measure
needs to be taken so that we can avoid heart disease and heart strokes. We examined the
relationship between one categorical response variable and various categorical independent
variables, so we chose Logistic Regression for prediction. To get better results, we can keep on
increasing various parameters and increase the size of the data.
TEAM CONTRIBUTION:
Everyone contributed equally as we helped each other when anyone of us was stuck on certain
point and that was the key highlight of our project. Our team-work led to a successful completion
of our project.

REFERENCES
 Heart Disease UCI. (2019). Kaggle.com. Retrieved 29 March 2019, from
https://www.kaggle.com/ronitf/heart-disease-uci/kernels
 ggplot2 histogram : Easy histogram graph with ggplot2 R package - Easy Guides - Wiki -
STHDA. (2019). Sthda.com. Retrieved 29 March 2019, from
http://www.sthda.com/english/wiki/ggplot2-histogram-easy-histogram-graph-with-
ggplot2-r-package
 A Short Introduction to the caret Package. (2019). Cran.r-project.org. Retrieved 29 March
2019, from https://cran.r-project.org/web/packages/caret/vignettes/caret.html
 Cui, B. (2019). Introduction to DataExplorer. Cran.r-project.org. Retrieved 29 March 2019,
from https://cran.r-project.org/web/packages/DataExplorer/vignettes/dataexplorer-
intro.html
 Machine Learning Part 3 : Logistic Regression. (2017). Towards Data Science. Retrieved
29 March 2019, from https://towardsdatascience.com/machine-learning-part-3-logistics-
regression-9d890928680f
 Evaluating Logistic Regression Models. (2015). R-bloggers. Retrieved 29 March 2019,
from https://www.r-bloggers.com/evaluating-logistic-regression-models/
 package, F., docs, R., & browser, R. (2019). theme_economist: ggplot color theme based
on the Economist in ggthemes: Extra Themes, Scales and Geoms for 'ggplot2'. Rdrr.io.
Retrieved 29 March 2019, from https://rdrr.io/cran/ggthemes/man/theme_economist.html