Introduction
The diabetes is threatening a lot of people nowadays, without having a perfect cure for it.There are actually two types of diabetes,namely Type 1 and Type 2.The type 2 diabetes is commonly called as diabetes mellitus.It can be defined as a chronic condition that affects the way the body processes blood sugar (glucose).We consider the mellitus here.After deep researches we found that, that some parameters are directly responsible to for the mellitus to occur.By using the data of the people with and without diabetes,a dataset has been build.We use that dataset to classify the people who are in the risk of getting diabetes.
Loading the required libraries
library(ggplot2)
library(ggvis)
library(corrplot)
library(caTools)
library(ROCR)Data Loading
The observations of the people are stored in a CSV format, named diabetes.csv.The data is loaded in the environment.Let’s check how the data is structured.
data = read.csv("C:/Users/crsri/Documents/Diabetes_Prediction/Data/diabetes.csv")
head(data)## Pregnancies Glucose BloodPressure SkinThickness Insulin BMI
## 1 6 148 72 35 0 33.6
## 2 1 85 66 29 0 26.6
## 3 8 183 64 0 0 23.3
## 4 1 89 66 23 94 28.1
## 5 0 137 40 35 168 43.1
## 6 5 116 74 0 0 25.6
## DiabetesPedigreeFunction Age Outcome
## 1 0.627 50 1
## 2 0.351 31 0
## 3 0.672 32 1
## 4 0.167 21 0
## 5 2.288 33 1
## 6 0.201 30 0summary(data)## Pregnancies Glucose BloodPressure SkinThickness
## Min. : 0.000 Min. : 0.0 Min. : 0.00 Min. : 0.00
## 1st Qu.: 1.000 1st Qu.: 99.0 1st Qu.: 62.00 1st Qu.: 0.00
## Median : 3.000 Median :117.0 Median : 72.00 Median :23.00
## Mean : 3.845 Mean :120.9 Mean : 69.11 Mean :20.54
## 3rd Qu.: 6.000 3rd Qu.:140.2 3rd Qu.: 80.00 3rd Qu.:32.00
## Max. :17.000 Max. :199.0 Max. :122.00 Max. :99.00
## Insulin BMI DiabetesPedigreeFunction Age
## Min. : 0.0 Min. : 0.00 Min. :0.0780 Min. :21.00
## 1st Qu.: 0.0 1st Qu.:27.30 1st Qu.:0.2437 1st Qu.:24.00
## Median : 30.5 Median :32.00 Median :0.3725 Median :29.00
## Mean : 79.8 Mean :31.99 Mean :0.4719 Mean :33.24
## 3rd Qu.:127.2 3rd Qu.:36.60 3rd Qu.:0.6262 3rd Qu.:41.00
## Max. :846.0 Max. :67.10 Max. :2.4200 Max. :81.00
## Outcome
## Min. :0.000
## 1st Qu.:0.000
## Median :0.000
## Mean :0.349
## 3rd Qu.:1.000
## Max. :1.000str(data)## 'data.frame': 768 obs. of 9 variables:
## $ Pregnancies : int 6 1 8 1 0 5 3 10 2 8 ...
## $ Glucose : int 148 85 183 89 137 116 78 115 197 125 ...
## $ BloodPressure : int 72 66 64 66 40 74 50 0 70 96 ...
## $ SkinThickness : int 35 29 0 23 35 0 32 0 45 0 ...
## $ Insulin : int 0 0 0 94 168 0 88 0 543 0 ...
## $ BMI : num 33.6 26.6 23.3 28.1 43.1 25.6 31 35.3 30.5 0 ...
## $ DiabetesPedigreeFunction: num 0.627 0.351 0.672 0.167 2.288 ...
## $ Age : int 50 31 32 21 33 30 26 29 53 54 ...
## $ Outcome : int 1 0 1 0 1 0 1 0 1 1 ...Correlations
The proportionalities of the attributes of the data can be identified by the correlation coefficient,either numerically or visually.They helps to know which attributes are highy dependent on the prediction variable:Outcome.
correlations <- cor(data)
correlations## Pregnancies Glucose BloodPressure
## Pregnancies 1.00000000 0.12945867 0.14128198
## Glucose 0.12945867 1.00000000 0.15258959
## BloodPressure 0.14128198 0.15258959 1.00000000
## SkinThickness -0.08167177 0.05732789 0.20737054
## Insulin -0.07353461 0.33135711 0.08893338
## BMI 0.01768309 0.22107107 0.28180529
## DiabetesPedigreeFunction -0.03352267 0.13733730 0.04126495
## Age 0.54434123 0.26351432 0.23952795
## Outcome 0.22189815 0.46658140 0.06506836
## SkinThickness Insulin BMI
## Pregnancies -0.08167177 -0.07353461 0.01768309
## Glucose 0.05732789 0.33135711 0.22107107
## BloodPressure 0.20737054 0.08893338 0.28180529
## SkinThickness 1.00000000 0.43678257 0.39257320
## Insulin 0.43678257 1.00000000 0.19785906
## BMI 0.39257320 0.19785906 1.00000000
## DiabetesPedigreeFunction 0.18392757 0.18507093 0.14064695
## Age -0.11397026 -0.04216295 0.03624187
## Outcome 0.07475223 0.13054795 0.29269466
## DiabetesPedigreeFunction Age Outcome
## Pregnancies -0.03352267 0.54434123 0.22189815
## Glucose 0.13733730 0.26351432 0.46658140
## BloodPressure 0.04126495 0.23952795 0.06506836
## SkinThickness 0.18392757 -0.11397026 0.07475223
## Insulin 0.18507093 -0.04216295 0.13054795
## BMI 0.14064695 0.03624187 0.29269466
## DiabetesPedigreeFunction 1.00000000 0.03356131 0.17384407
## Age 0.03356131 1.00000000 0.23835598
## Outcome 0.17384407 0.23835598 1.00000000corrplot(correlations, method="color")
Visualization
Visualizations are used to grasp the structure of data and its relations,like how they vary and their relationships with the otehr data.They are said to be EDA.
A matrix of scatterplots is produce for this dataset.
pairs(data, col=data$Outcome)
Glucose and Insulin
The glucose and the insulin are the major factors of the diabetes…which in turn have direct proportionality in the future during the diabetes.They are the major cause of the occurence.They are strong correlated on each other.
data %>% ggvis(~Glucose,~Insulin,fill =~Outcome) %>% layer_points()BMI ad DiabetesPedigreeFunction
The BMI and DiabetesPedigreeFunction is plotted here.
data %>% ggvis(~BMI,~DiabetesPedigreeFunction,fill =~Outcome) %>% layer_points()Age and Pregnancies
The males have 0 for the pregnancy attribute, which is why we find a lot of values plottinh zero in this grpah.
data %>% ggvis(~Age,~Pregnancies,fill =~Outcome) %>% layer_points()Preparing the data
The dataset is divided as two parts, training data and testing data, with a Splitratio of 0.75. It means that 2/3rds of the data is labelled by training set and the rest 1/3rd of data is the testing set.The division of the dataset is by means of a random order generated by the seed.
set.seed(88)
split <- sample.split(data$Outcome, SplitRatio = 0.75)
data_train <- subset(data, split == TRUE)
data_test <- subset(data, split == FALSE)Logistic regression
The Logistic regression helps to classify the concern person will get diabetes or not.Since we are using the logistic regression we have to mention that, family = binomial.We are using all the attributes we have in the dataset.Let us take a look at the summary.
model <- glm (Outcome ~ .-Pregnancies + Glucose + BloodPressure + SkinThickness + Insulin + BMI + DiabetesPedigreeFunction + Age, data = data_train, family = binomial)
summary(model)##
## Call:
## glm(formula = Outcome ~ . - Pregnancies + Glucose + BloodPressure +
## SkinThickness + Insulin + BMI + DiabetesPedigreeFunction +
## Age, family = binomial, data = data_train)
##
## Deviance Residuals:
## Min 1Q Median 3Q Max
## -2.4254 -0.7250 -0.4361 0.7487 2.9829
##
## Coefficients:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) -8.3339721 0.8159489 -10.214 < 2e-16 ***
## Glucose 0.0382162 0.0044235 8.639 < 2e-16 ***
## BloodPressure -0.0088309 0.0060059 -1.470 0.1415
## SkinThickness 0.0007624 0.0081902 0.093 0.9258
## Insulin -0.0017095 0.0010823 -1.580 0.1142
## BMI 0.0792632 0.0169318 4.681 2.85e-06 ***
## DiabetesPedigreeFunction 0.7386714 0.3332368 2.217 0.0266 *
## Age 0.0204344 0.0095270 2.145 0.0320 *
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## (Dispersion parameter for binomial family taken to be 1)
##
## Null deviance: 745.11 on 575 degrees of freedom
## Residual deviance: 552.82 on 568 degrees of freedom
## AIC: 568.82
##
## Number of Fisher Scoring iterations: 5Prediction
The trained model is used to predict the data for the testing data and for the training data(For checking accuracy purposes and for ROC curve)
predict_train <- predict(model, type = 'response')
predict_test <- predict(model, newdata = data_test, type = 'response')ROC Curve
ROCRpred <- prediction(predict_train, data_train$Outcome)
ROCRperf <- performance(ROCRpred, 'tpr','fpr')
plot(ROCRperf, colorize = TRUE, text.adj = c(-0.2,1.7))
Comparison
By comparing the real values with the real data, we can see the how our machine learning algorithm performs.
predict_test_c = predict_test
i = 1
while(i <= length(predict_test))
{
if(predict_test[i] < 0.5)
predict_test_c[i] = 0
else
predict_test_c[i] = 1
i = i + 1;
}
compare <- data.frame(data_test$Outcome,predict_test_c)
colnames(compare) <- c("Observed Values","Predicted values")
ggplot(data = compare,aes(x = "Observed Values", y = "Predicted values")) + geom_abline() +
xlab("Observed Values") + ylab("Predicted values") + theme_classic()
compare## Observed Values Predicted values
## 6 0 0
## 8 0 0
## 9 1 1
## 10 1 0
## 11 0 0
## 14 1 1
## 16 1 0
## 32 1 1
## 33 0 0
## 37 0 0
## 38 1 0
## 41 0 1
## 44 1 1
## 45 0 1
## 46 1 1
## 49 1 0
## 56 0 0
## 60 0 0
## 62 1 0
## 70 0 0
## 73 1 1
## 77 0 0
## 78 0 0
## 84 0 0
## 86 0 0
## 89 1 1
## 91 0 0
## 93 0 0
## 94 1 0
## 95 0 0
## 102 0 0
## 103 0 0
## 105 0 0
## 110 1 0
## 111 1 1
## 114 0 0
## 124 0 0
## 128 0 0
## 130 1 0
## 142 0 0
## 143 0 0
## 150 0 0
## 153 1 1
## 163 0 0
## 164 0 0
## 168 0 0
## 182 0 0
## 192 0 0
## 195 0 0
## 199 1 0
## 201 0 0
## 204 0 0
## 209 0 0
## 216 1 1
## 219 1 0
## 225 0 0
## 227 0 0
## 228 1 1
## 236 1 1
## 239 1 1
## 240 0 0
## 244 1 0
## 256 1 0
## 262 1 1
## 264 0 1
## 272 0 0
## 280 0 0
## 281 1 1
## 283 0 0
## 285 1 0
## 291 0 0
## 292 1 0
## 299 1 0
## 304 1 1
## 312 0 0
## 315 1 0
## 323 1 0
## 326 0 0
## 327 1 0
## 341 0 0
## 342 0 0
## 343 0 0
## 344 0 0
## 346 0 0
## 350 1 0
## 356 1 1
## 357 1 0
## 358 1 1
## 363 0 0
## 364 1 1
## 367 1 0
## 374 0 0
## 379 1 1
## 381 0 0
## 382 0 0
## 388 1 0
## 391 0 0
## 392 1 1
## 395 1 1
## 396 0 0
## 408 0 0
## 414 0 0
## 417 0 0
## 419 0 0
## 422 0 0
## 424 0 0
## 431 0 0
## 432 0 0
## 433 0 0
## 436 1 1
## 437 0 1
## 439 0 0
## 448 0 0
## 449 1 0
## 450 0 0
## 451 0 0
## 453 0 0
## 456 1 1
## 463 0 0
## 466 0 0
## 473 0 0
## 478 0 0
## 493 0 0
## 498 0 0
## 500 0 1
## 504 0 0
## 508 0 0
## 509 0 0
## 513 0 0
## 531 0 0
## 532 0 0
## 533 0 0
## 536 1 1
## 538 0 0
## 542 1 0
## 543 1 0
## 548 0 0
## 550 0 1
## 562 1 1
## 563 0 0
## 567 0 0
## 573 0 0
## 577 0 0
## 580 1 1
## 583 0 0
## 585 1 0
## 586 0 0
## 592 0 0
## 599 1 1
## 606 0 0
## 608 0 0
## 610 0 0
## 623 0 1
## 625 0 0
## 627 0 0
## 636 1 0
## 639 1 0
## 640 0 0
## 652 0 0
## 655 0 0
## 663 1 1
## 664 1 1
## 665 1 0
## 671 0 1
## 673 0 0
## 675 0 0
## 680 0 0
## 681 0 0
## 691 0 0
## 694 1 1
## 695 0 0
## 700 0 1
## 703 1 1
## 711 0 0
## 714 0 0
## 719 0 0
## 721 0 0
## 724 0 0
## 728 0 0
## 736 0 0
## 740 1 0
## 741 1 1
## 744 1 1
## 746 0 0
## 747 1 1
## 749 1 1
## 753 0 0
## 757 0 0
## 759 0 0
## 760 1 1
## 764 0 0
## 766 0 0Conclusion
The results can be improved by applyting the feature scaling and data cleaning.From this project we predict the type 2 diabetes, commonly called as diabetes mellitus.As a result it can help to improve their health conditions.
Things to do in future
Data cleaning and Feature Scaling have to be done with the data.Then running the prepared data with the logistic regression to get the improved results.