-
Notifications
You must be signed in to change notification settings - Fork 0
/
A&E_Challenge_Kelechi.R
224 lines (166 loc) · 7.3 KB
/
A&E_Challenge_Kelechi.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
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
#Code to explore A&E data and predict whether patients have been admitted.
# Install relevant packages
#install.packages('caret','tidyverse','caTools')
# install.packages("ggcorrplot", "caTools")
# install.packages('caTools', repos = "http://cran.us.r-project.org")
# install.packages("httpgd", repos = "http://cran.us.r-project.org")
# install.packages("languageserver", repos = "http://cran.us.r-project.org")
# install.packages('skimr', repos = "http://cran.us.r-project.org")
# install.packages("corrplot", repos = "http://cran.us.r-project.org")
# install.packages("fastDummies", repos = "http://cran.us.r-project.org")
# install.packages("lubridate", repos = "http://cran.us.r-project.org")
# Load required packages
library(tidyverse)
library(caTools)
library(caret)
library(dplyr)
library(skimr)
library(corrplot)
library(fastDummies)
library(lubridate)
#A. Import all relevant files & read into dataframes
#download.file(url = "",destfile = "*.csv")
file_path <- "bip-ae-technical-challenge/training_set.csv"
train_data <- read.csv(file_path)
test_path <- "bip-ae-technical-challenge/test_set.csv"
test_data <- read.csv(test_path)
#Task 1. Explore the datasets
# Checking dimensions. Shows number of columns/features. Makes sense for test to gave one less.
dim(train_data)
dim(test_data)
# Summary statistics (str = structure)
str(train_data)
# Name / type / first 10 values
summary(train_data)
# Visualize variable distributions
# Histogram for numeric variables
num_col <- train_data %>% select(where(is.numeric))
str(num_col)
for (i in 1:ncol(num_col)) {
hist(num_col[[i]],
main = paste("Histogram of", colnames(num_col)[i]),
col = "blue", ylim = c(0, 80000))
}
# Bar plots for categorical variables
var_col <- train_data %>% select(!where(is.numeric))
str(var_col)
#1. Patient Admittance
table(train_data$Admitted_Flag)
ggplot(train_data, aes(x = Admitted_Flag)) +
geom_histogram(bins=30, fill = "magenta")+
labs(title = "Distribution of Patient Admittance",
x = "Admitted_Flag",
y = "Number of patients")
## Correlation between variables
# Correlation matrix for numeric columns
num_col <- train_data %>% select(where(is.numeric))
summary(num_col)
# Display the correlation matrix as a heatmap
cor_matrix <- cor(num_col, use = "pairwise.complete.obs")
corrplot(cor_matrix, method = "color")
#Task 2. Create a validation dataset
# setting seed to generate a reproducible random sampling
set.seed(100)
# dividing the dataset into an 80:20 ratio
spl <- createDataPartition(train_data$Admitted_Flag, p = 0.8, list = FALSE)
# selecting part of dataset which belongs to the 80%
train <- train_data[spl, ]
# selecting part of dataset which belongs to the 20%
test <- train_data[-spl, ]
# Store X and Y for later use.
x <- train %>% select(-Admitted_Flag)
y <- train$Admitted_Flag
str(x)
str(y)
# overview of training data
skimmed <- skim(train)
skimmed
#B. Preprocess the training data
# Count missing values in the training set
sum(is.na(train))
# Remove columns with excessive number of NA values
thresh <- 0.5 # Define threshold for NA values
#Identify columns in training set with more than 50% NA
colnames(train[which(colMeans(is.na(train)) > thresh)])
# drop columns with more than 50% NA
train <- train[, which(colMeans(!is.na(train)) > thresh)]
# data imputation for missing values
# Set any null values in the 'Sex' column of train to 0
train$Sex[is.na(train$Sex)] <- 0
# Set any null values in the 'Provider_Patient_Distance_Miles' column of train to the mean of the column
train$Provider_Patient_Distance_Miles[is.na(train$Provider_Patient_Distance_Miles)] <- mean(train$Provider_Patient_Distance_Miles, na.rm = TRUE)
test$Provider_Patient_Distance_Miles[is.na(test$Provider_Patient_Distance_Miles)] <- mean(test$Provider_Patient_Distance_Miles, na.rm = TRUE)
# Set any null values in the 'IMD_Decile_From_LSOA' column of train to 5
train$IMD_Decile_From_LSOA[is.na(train$IMD_Decile_From_LSOA)] <- 5
test$IMD_Decile_From_LSOA[is.na(test$IMD_Decile_From_LSOA)] <- 5
# # Set any null values of 'Length_Of_Stays_Days' to 0
# train$Length_Of_Stay_Days[is.na(train$Length_Of_Stay_Days)] <- 0
# test$Length_Of_Stay_Days[is.na(test$Length_Of_Stay_Days)] <- 0
# Replace 'NA' in "EA_HRG" with the value "Nothing"
train$AE_HRG[is.na(train$AE_HRG)] <- "Nothing"
test$AE_HRG[is.na(test$AE_HRG)] <- "Nothing"
# create One-Hot Encoding(dummy variables) and ordinal encoding for categorical variables
age_band_levels <- c("1-17", "18-24", "25-44", "45-64", "65-84", "85")
train$Age_Band <- as.integer(factor(train$Age_Band, levels = age_band_levels, ordered = TRUE))
test$Age_Band <- as.integer(factor(test$Age_Band, levels = age_band_levels, ordered = TRUE))
train <- dummy_cols(train, select_columns = "AE_HRG", remove_first_dummy = TRUE, remove_selected_columns = TRUE)
test <- dummy_cols(test, select_columns = "AE_HRG", remove_first_dummy = TRUE, remove_selected_columns = TRUE)
# Normalize data formats
# Load required library
# Convert 'AE_Arrive_Date' to datetime format
train$AE_Arrive_Date <- as.POSIXct(train$AE_Arrive_Date)
test$AE_Arrive_Date <- as.POSIXct(test$AE_Arrive_Date)
# Extract date components
train <- train %>%
mutate(Arrival_Year = year(AE_Arrive_Date),
Arrival_Month = month(AE_Arrive_Date),
Arrival_Day = day(AE_Arrive_Date),
Arrival_DayOfWeek = wday(AE_Arrive_Date, label = FALSE) - 1) # Monday=0, Sunday=6
test <- test %>%
mutate(Arrival_Year = year(AE_Arrive_Date),
Arrival_Month = month(AE_Arrive_Date),
Arrival_Day = day(AE_Arrive_Date),
Arrival_DayOfWeek = wday(AE_Arrive_Date, label = FALSE) - 1) # Monday=0, Sunday=6
# Drop the original 'AE_Arrive_Date' field
train <- train %>% select(-AE_Arrive_Date)
test <- test %>% select(-AE_Arrive_Date)
#C. Reanalyse the new training set
# Summary Statistics
str(train)
# Missing values analysis
# Check missing values after data imputation
sum(is.na(train))
colSums(is.na(train))
# Visualize importance of variables using featurePlot()
featurePlot(x = train[, -which(names(train) %in% "Admitted_Flag")],
y = train$target,
plot = "density",
scales = list(x = list(relation = "free"),
y = list(relation = "free")))
# Feature selection using recursive feature elimination(rfe)
# Check for NA values in the predictors
sum(is.na(train[, -which(names(train) == "Admitted_Flag")]))
# Say goodbye to 10285 rows with NA values. rip.
train <- train[complete.cases(train[, -which(names(train) == "Admitted_Flag")]), ]
# Sample a quarter of the data
sample_size <- nrow(train) * 0.2
train_subset <- train %>% sample_n(sample_size)
ctrl <- rfeControl(functions = rfFuncs, method = "cv", number = 10)
results <- rfe(train_subset[, -which(names(train_subset) == "Admitted_Flag")],
train_subset$Admitted_Flag,
sizes=c(1:ncol(train_subset)-1),
rfeControl=ctrl)
print(results)
# The top 5 variables (out of 8):
# AE_Time_Mins, AE_Num_Investigations, AE_Arrival_Mode, Age_Band, AE_HRG_Nothing
#E. Evaluate ML algorithms
#1. Build models
#2. Select best model
#3. Compare accuracy using visualizations
#4. Summarise best model
#5. rank features by importance
#F. Make predictions using test data and the most accurate model.
#G. Create ensemble predictions for improved accuracy (Stretch goal)
# - Bagging
# - Boosting
# - Stacking