Machine_Learning for prediction of diabetes

Zhanyou Xu
August 25, 2018

To test the app, please visit: https://bioinformatics-biostatistics.shinyapps.io/AAA111/.

To access the R codes and esting data at Github: https://github.com/xuzhanyouiowa/develop_data_products

Instructions to run the App

The app includes the following four pains

Load your own data: Please download the testing data “New_data.csv” from Github to test the App, the like is: https://github.com/xuzhanyouiowa/develop_data_products/blob/gh-pages/New_data.csv
Model selection:
- 2 models: 1) Random Forest (RF), 2) Support Vector Machine (SVM)
Cross validation and different percentage of training/testing data sets
Run the prediction and download the results

Very import notes:

as for any machine learning prediction, the data format is very important. This App use data “PimaIndiansDiabetes” from R package “mlbench”. You can load the library and use the data to test this app. Please keep the same number of cloumns and same column names.

library(mlbench)
data(PimaIndiansDiabetes)
# show the top rows of the data
PimaIndiansDiabetes[1:2,]

  pregnant glucose pressure triceps insulin mass pedigree age diabetes
1        6     148       72      35       0 33.6    0.627  50      pos
2        1      85       66      29       0 26.6    0.351  31      neg

Visualize the correlation among the predictiors with correlation plot

library("corrplot")
M <- cor(PimaIndiansDiabetes[, 1:(ncol(PimaIndiansDiabetes)-1)])
corrplot(M, method = "circle")

plot of chunk unnamed-chunk-2

Slide With 5-fold cross validation

    folder accuracy
1 folder_1   0.7922
2 folder_2   0.7727
3 folder_3   0.8182
4 folder_4   0.7468
5 folder_5   0.6908

[1] "Thank you for your attention, and please try my Shiny App at:        <https://bioinformatics-biostatistics.shinyapps.io/AAA111/>, and email me at        zhanyouxu@gmail.com if you have any questions and feedbacks. Thank you!"