-
Notifications
You must be signed in to change notification settings - Fork 1
/
Run (final).R
200 lines (184 loc) · 8.48 KB
/
Run (final).R
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
library(caret)
library(e1071)
library(rpart)
library(rattle)
library(rpart.plot)
library(RColorBrewer)
library(randomForest)
library(party)
library(RWeka)
library(elmNN)
library(nnet)
library(neuralnet)
library(sampling)
### FUNCTIONS ###
## Print output to results.txt file in current working directory
out <- function(x)
{
cat(date(), ": ", x, "\n", file="output.txt", append=TRUE)
}
## Provide some basic stats about an attribute
explore <- function(x)
{
print(summary(x))
dist <- factor(x, exclude=NULL)
print(table(dist))
plot(table(dist))
}
## Run analysis on a dataset and testset
analyze <- function(dataset, testset, model=c("svm", "rforest", "nnet"), seed, formula, na.remove=TRUE)
{
if (na.remove == TRUE)
{
dataset <- dataset[complete.cases(dataset),]
}
# Need to set seed again for the model
set.seed(seed)
if (model == "svm")
{
svm_fit <- svm(formula, data=dataset, na.action=na.omit)
testset$Predicted <- predict(svm_fit, testset)
}
if (model == "rforest")
{
rforest_fit <- randomForest(formula, data=dataset, importance=TRUE, ntree=1000)
#varImpPlot(rforest_fit)
testset$Predicted <- predict(rforest_fit, testset)
}
if (model == "nnet")
{
nnet_fit <- avNNet(formula, data=dataset, repeats=3, bag=FALSE, allowParallel=TRUE, decay=0.001, size=5)
testset$Predicted <- predict(nnet_fit, testset, type="class")
}
matrix <- confusionMatrix(testset$TreatmentComplete, testset$Predicted)
accuracy <- round(matrix$overall[1] * 100, 2)
cols <- paste (colnames(dataset), collapse=",")
str <- paste ("Model", model, "Columns", cols, "Accuracy", accuracy)
accuracy
}
# Get Precision/Recall/Accuracy table
getresults <- function(x, y)
{
matrix <- confusionMatrix(x, y)
accuracy <- matrix$overall[1] # Correctness of model
precision <- matrix$byClass[3] # Positive prediction value
neg_precision <- matrix$byClass[4] # Negative prediction value
sensitivity <- matrix$byClass[1] # True positive recognition rate (aka recall)
specificity <- matrix$byClass[2] # True negative recognition rate
type1_error <- 0 # FP
type2_error <- 0 # FN
results <- c(accuracy, precision, sensitivity, specificity)
results
}
#################
attributes <- c("PatientID", "TreatmentComplete", "Gender", "AgeGroup", "WeightGroup", "HeightGroup", "MaritalStatus", "Religion", "Caste",
"ScreeningYear", "RegistrationYear", "Fever", "Cough", "CoughDuration", "ProductiveCough", "BloodInCough", "NightSweats", "WeightLoss", "TBHistory", "TBInFamily", "SeverityScore",
"RegistrationDelay", "SmearTested", "SputumResultDelay", "SmearResult", "SmearPositive", "ScreeningToSmearDelay", "XRayDone", "XRayResultDelay", "XRayResults", "XRayIndicative", "ScreeningToXRayDelay", "GeneXpertTested", "GeneXpertResult", "DrugResistance", "GXPPositive",
"ScreeningToGXPDelay", "DiagnosisDone", "ScreeningToDiagnosisDelay", "DiagnosedBy", "DiagnosisAntibiotic", "TBSymptomsDiagnosed", "TBContactDiagnosed", "Diagnosis", "LargeLymphDiagnosed", "LymphBiopsyDiagnosed", "MantouxDiagnosed", "PastTBDiagnosed", "XRaySuggestiveDiagnosed",
"ScreeningToBaselineDelay", "SmearToBaselineDelay", "XRayToBaselineDelay", "GXPToBaselineDelay", "DiagnosisToBaselineDelay", "BaselineWeightGroup", "BaselinePatientCategory", "BaselinePatientType", "BaselineRegimen", "BaselineDoseCombination", "BaselineStreptomycin", "ScreeningToBaselineWeightDifference", "BaselineWeightGroup", "HasTreatmentSupporter", "DiseaseCategory", "DiseaseSite", "DoseCombination")
seeds <- c(10, 300, 5000, 700)
setwd("D:/Datasets/Tuberculosis")
limits <- c(0.2, 0.5, 1)
test_cases <- 0.2
prune_columns <- FALSE
sampling <- "stratified" # random or stratified
out(paste("Sampling:", sampling))
for (limit in limits)
{
out(paste("Limit:", limit * 100))
for (seed in seeds)
{
out(paste("Seed:", seed))
# Read the dataset fresh
dt <- read.csv("dataset_clean.csv", stringsAsFactors=TRUE)
# Remove DIED and limit the dataset only to the columns till baseline treatment
dt <- dt[dt$TreatmentOutcome != 'DIED', attributes]
# Limit to given percentage of data
set.seed(seed)
dt <- dt[sample(1:nrow(dt), nrow(dt) * limit, replace=FALSE), ]
# Separate test data initially
set.seed(seed)
if (sampling == "random")
{
indices <- sample(1:nrow(dt), nrow(dt)*test_cases, replace=FALSE)
} else if (sampling == "stratified")
{
indices <- strata(data=dt, method="srswor", size=nrow(dt)*test_cases)$ID_unit
}
test <- dt[indices,]
train <- dt[-indices,]
# For every individual column, calculate the accuracy and pick the column with highest accuracy
if (prune_columns)
{
col_length <- length(colnames(dt))
model="svm"
accuracies <- data.frame(Col=colnames(dt)[3:col_length], Acc=rep(0,col_length-2))
for (i in 3:col_length)
{
dataset <- dt[,c(1,2,i)]
formula <- as.factor(TreatmentComplete) ~ .
accuracy <- analyze (dataset, test, model, seed, formula)
accuracies$Acc[accuracies$Col == colnames(dt)[i]] <- accuracy
}
svm_accuracies <- accuracies
model="rforest"
accuracies <- data.frame(Col=colnames(dt)[3:col_length], Acc=rep(0,col_length-2))
for (i in 3:col_length)
{
dataset <- dt[,c(1,2,i)]
formula <- as.factor(TreatmentComplete) ~ .
accuracy <- analyze (dataset, test, model, seed, formula)
accuracies$Acc[accuracies$Col == colnames(dt)[i]] <- accuracy
}
rforest_accuracies <- accuracies
model="nnet"
accuracies <- data.frame(Col=colnames(dt)[3:col_length], Acc=rep(0,col_length-2))
for (i in 3:col_length)
{
dataset <- dt[,c(1,2,i)]
formula <- as.factor(TreatmentComplete) ~ .
accuracy <- analyze (dataset, test, model, seed, formula)
accuracies$Acc[accuracies$Col == colnames(dt)[i]] <- accuracy
}
nnet_accuracies <- accuracies
# Feature selection. For each model, keep appending every feature and recalculate accuracy
svm_accuracies
rforest_accuracies
nnet_accuracies
yes_prob <- length(dt$TreatmentComplete[dt$TreatmentComplete == 'YES'])/nrow(dt) * 100
# The columns with accuracy less than probability of YES on all the models tried are of little use, because hypothetically, they add nothing to prediction
accuracies <- data.frame(attribute=svm_accuracies$Col, svm=svm_accuracies$Acc, rforest=rforest_accuracies$Acc, nnet=nnet_accuracies$Acc)
# Get the columns indices that are at least as accurate as yes_prob
attributes <- c()
for (i in 1:nrow(accuracies))
{
# If any one model gives higher accuracy, then keep it
if (accuracies[i,"svm"] >= yes_prob || accuracies[i,"rforest"] >= yes_prob || accuracies[i,"nnet"] >= yes_prob)
{
attributes <- c(attributes, accuracies$attribute[i])
}
}
print(sort(attributes))
train <- train[,attributes]
test <- test[,attributes]
}
formula <- TreatmentComplete ~ .
# Run with default parameters
svm_fit <- svm(formula, data=train, na.action=na.exclude)
test$Predicted <- predict(svm_fit, test)
svm_results <- getresults(test$Predicted, test$TreatmentComplete)
rforest_fit <- randomForest(formula, data=train, importance=TRUE)
test$Predicted <- predict(rforest_fit, test)
rforest_results <- getresults(test$Predicted, test$TreatmentComplete)
nnet_fit <- avNNet(formula, data=train, allowParallel=TRUE, size=5, repeats=10, decay=0.001, bag=FALSE, MaxNWts=5000, trace=FALSE)
test$Predicted <- predict(nnet_fit, test, type="class")
matrix <- confusionMatrix(test$TreatmentComplete, test$Predicted)
nnet_results <- getresults(test$TreatmentComplete, test$Predicted)
print(svm_results)
print(rforest_results)
print(nnet_results)
out(svm_results)
out(rforest_results)
out(nnet_results)
}
}