first baby steps in building a predictive model

This commit is contained in:
AG Damsbo 2021-11-12 08:11:16 +01:00
parent 364bbe290e
commit ee38f4ce89
3 changed files with 341 additions and 0 deletions

99
1 PA Decline/dataset.R Normal file
View File

@ -0,0 +1,99 @@
# Data
## Import from previous work
dta<-read.csv("/Volumes/Data/exercise/source/background.csv",na.strings = c("NA","","unknown"),colClasses = "character")
## Cleaning and enhancing
dta$pase_drop<-factor(ifelse((dta$pase_0_q=="q_2"|dta$pase_0_q=="q_3"|dta$pase_0_q=="q_4")&dta$pase_06_q=="q_1","yes","no"),levels = c("no","yes"))
dta$pase_drop[is.na(dta$pase_6)]<-NA
dta$pase_drop[is.na(dta$pase_0)]<-NA
## Selection of data set and formatting
library(dplyr)
dta_f<-dta %>% filter(pase_0_q != "q_1" & !is.na(pase_drop))
variable_names<-c("age","sex","weight","height",
"bmi",
"smoke_ever",
"civil",
"diabetes",
"hypertension",
"pad",
"afli",
"ami",
"tci",
"nihss_0",
"thrombolysis",
"thrombechtomy",
"rep_any","pase_0_q","pase_drop")
library(daDoctoR)
dta2<-dta_f[,variable_names]
dta2<-col_num(c("age","weight","height","bmi","nihss_0"),dta2)
dta2<-col_fact(c("sex","smoke_ever","civil","diabetes", "hypertension","pad", "afli", "ami", "tci","thrombolysis", "thrombechtomy","rep_any","pase_0_q","pase_drop"),dta2)
## Partitioning
library(caret)
set.seed(100)
## Step 1: Get row numbers for the training data
trainRowNumbers <- createDataPartition(dta2$pase_drop, p=0.8, list=FALSE)
## Step 2: Create the training dataset
trainData <- dta2[trainRowNumbers,]
## Step 3: Create the test dataset
testData <- dta2[-trainRowNumbers,]
y_test = testData[,"pase_drop"]
# Store X and Y for later use.
x = trainData %>% select(!matches("pase_drop"))
y = trainData[,"pase_drop"]
# Normalization and dummy binaries
# One-Hot Encoding
# Creating dummy variables is converting a categorical variable to as many binary variables as here are categories.
dummies_model <- dummyVars(pase_drop ~ ., data=trainData)
# Create the dummy variables using predict. The Y variable (Purchase) will not be present in trainData_mat.
trainData_mat <- predict(dummies_model, newdata = trainData)
# # Convert to dataframe
trainData <- data.frame(trainData_mat)
# # See the structure of the new dataset
str(trainData)
dummies_model <- dummyVars(pase_drop ~ ., data=testData)
testData_mat <- predict(dummies_model, newdata = testData)
testData <- data.frame(testData_mat)
preProcess_range_model <- preProcess(testData, method='range')
testData <- predict(preProcess_range_model, newdata = testData)
testData$pase_drop<-y_test
# Imputation
library(RANN) # required for knnInpute
preProcess_missingdata_model <- preProcess(trainData, method='knnImpute')
# preProcess_missingdata_model
trainData <- predict(preProcess_missingdata_model, newdata = trainData) # Giver fejl??
anyNA(trainData)
# skimr::skim(trainData)
# skimr::skim(x)
preProcess_range_model <- preProcess(trainData, method='range')
trainData <- predict(preProcess_range_model, newdata = trainData)
# Append the Y variable
trainData$pase_drop <- y
# Export
write.csv(trainData,"/Users/au301842/PhysicalActivityandStrokeOutcome/data/trainData.csv",row.names = FALSE)
write.csv(testData,"/Users/au301842/PhysicalActivityandStrokeOutcome/data/testData.csv",row.names = FALSE)

View File

@ -0,0 +1,142 @@
# https://www.machinelearningplus.com/machine-learning/caret-package/
# install.packages(c('caret', 'skimr', 'RANN', 'randomForest', 'fastAdaboost', 'gbm', 'xgboost', 'caretEnsemble', 'C50', 'earth'))
# Load the caret package
library(caret)
# Import dataset
orange <- read.csv('https://raw.githubusercontent.com/selva86/datasets/master/orange_juice_withmissing.csv')
# Structure of the dataframe
str(orange)
# See top 6 rows and 10 columns
head(orange[, 1:10])
# Create the training and test datasets
set.seed(100)
# Step 1: Get row numbers for the training data
trainRowNumbers <- createDataPartition(orange$Purchase, p=0.8, list=FALSE)
# Step 2: Create the training dataset
trainData <- orange[trainRowNumbers,]
# Step 3: Create the test dataset
testData <- orange[-trainRowNumbers,]
# Store X and Y for later use.
x = trainData[, 2:18]
y = trainData$Purchase
library(skimr)
skimmed <- skim(trainData)
skimmed
# Create the knn imputation model on the training data
preProcess_missingdata_model <- preProcess(trainData, method='knnImpute')
preProcess_missingdata_model
# Use the imputation model to predict the values of missing data points
library(RANN) # required for knnInpute
trainData <- predict(preProcess_missingdata_model, newdata = trainData)
anyNA(trainData)
# One-Hot Encoding
# Creating dummy variables is converting a categorical variable to as many binary variables as here are categories.
dummies_model <- dummyVars(Purchase ~ ., data=trainData)
# Create the dummy variables using predict. The Y variable (Purchase) will not be present in trainData_mat.
trainData_mat <- predict(dummies_model, newdata = trainData)
# # Convert to dataframe
trainData <- data.frame(trainData_mat)
# # See the structure of the new dataset
str(trainData)
preProcess_range_model <- preProcess(trainData, method='range')
trainData <- predict(preProcess_range_model, newdata = trainData)
# Append the Y variable
trainData$Purchase <- y
apply(trainData[, 1:10], 2, FUN=function(x){c('min'=min(x), 'max'=max(x))})
featurePlot(x=trainData[,1:18],
y=factor(trainData$Purchase),
plot="box",
strip=strip.custom(par.strip.text=list(cex=.7)),
scales = list(x = list(relation="free"),
y = list(relation="free")))
featurePlot(x=trainData[,1:18],
y=factor(trainData$Purchase),
plot="density",
strip=strip.custom(par.strip.text=list(cex=.7)),
scales = list(x = list(relation="free"),
y = list(relation="free")))
# 5
set.seed(100)
options(warn=-1)
subsets <- c(1:5, 10, 15, 18)
ctrl <- rfeControl(functions = rfFuncs,
method = "repeatedcv",
repeats = 5,
verbose = FALSE)
lmProfile <- rfe(x=trainData[, 1:18], y=factor(trainData$Purchase),
sizes = subsets,
rfeControl = ctrl)
lmProfile
# See available algorithms in caret
modelnames <- dput(names(getModelInfo()))
# modelnames <- paste(names(getModelInfo()), collapse=', ')
modelnames
# Set the seed for reproducibility
set.seed(100)
# Train the model using randomForest and predict on the training data itself.
model_mars = train(Purchase ~ ., data=trainData, method='earth')
fitted <- predict(model_mars)
model_mars
plot(model_mars, main="Model Accuracies with MARS")
varimp_mars <- varImp(model_mars)
plot(varimp_mars, main="Variable Importance with MARS")
## 6.4
# Step 1: Impute missing values
testData2 <- predict(preProcess_missingdata_model, testData)
# Step 2: Create one-hot encodings (dummy variables)
testData3 <- predict(dummies_model, testData2)
# Step 3: Transform the features to range between 0 and 1
testData4 <- predict(preProcess_range_model, testData3)
# View
head(testData4[, 1:10])
predicted <- predict(model_mars, testData4)
head(predicted)
# Compute the confusion matrix
confusionMatrix(reference = factor(testData$Purchase), data = predicted, mode='everything', positive='MM')

View File

@ -0,0 +1,100 @@
---
title: "predictive_model"
output: pdf_document
---
```{r setup, include=FALSE}
knitr::opts_chunk$set(echo = TRUE)
```
# Data
```{r}
library(caret)
library(pROC)
library(daDoctoR)
```
Import
```{r}
trainData<-read.csv("/Users/au301842/PhysicalActivityandStrokeOutcome/data/trainData.csv",)
testData<-read.csv("/Users/au301842/PhysicalActivityandStrokeOutcome/data/testData.csv",)
```
# Prediction
Inspiration: https://stackoverflow.com/questions/30366143/how-to-compute-roc-and-auc-under-roc-after-training-using-caret-in-r and https://www.machinelearningplus.com/machine-learning/caret-package/
## Early visualisation
```{r}
featurePlot(x = trainData %>% select(!matches("pase_drop")),
y = factor(trainData$pase_drop),
plot = "box",
strip=strip.custom(par.strip.text=list(cex=.7)),
scales = list(x = list(relation="free"),
y = list(relation="free")))
featurePlot(x = trainData %>% select(!matches("pase_drop")),
y = factor(trainData$pase_drop),
plot = "density",
strip=strip.custom(par.strip.text=list(cex=.7)),
scales = list(x = list(relation="free"),
y = list(relation="free")))
```
```{r}
subsets <- c(1:10, 15, 18,33)
ctrl <- rfeControl(functions = rfFuncs,
method = "repeatedcv",
repeats = 5,
verbose = FALSE)
lmProfile <- rfe(x = trainData %>% select(!matches("pase_drop")),
y = trainData$pase_drop,
sizes = subsets,
rfeControl = ctrl)
lmProfile
```
```{r}
set.seed(1000)
forest.model <- train(pase_drop ~., trainData)
result.predicted.prob <- predict(forest.model, testData, type="prob") # Prediction
result.roc <- roc(factor(testData$pase_drop), result.predicted.prob$no) # Draw ROC curve.
plot(result.roc, print.thres="best", print.thres.best.method="closest.topleft")
result.coords <- coords(result.roc, "best", best.method="closest.topleft", ret=c("threshold", "accuracy"))
print(result.coords)#to get threshold and accuracy
```
```{r}
library(MLeval)
myTrainingControl <- trainControl(method = "cv",
number = 10,
savePredictions = TRUE,
classProbs = TRUE,
verboseIter = TRUE)
randomForestFit = train(x = trainData[,1:32],
y = as.factor(trainData$pase_drop),
method = "rf",
trControl = myTrainingControl,
preProcess = c("center","scale"),
ntree = 50)
x <- evalm(randomForestFit)
x$roc
x$stdres
```