Appendix

R Codes:

data <- Worli_Mumbai_PM2_5

head(data)
summary(data)
colSums(is.na(data))
missing_percent <- colSums(is.na(data)) / nrow(data) * 100
missing_percent
#missing percent in original data = 7.68

complete_data <- data[complete.cases(data),]

head(complete_data)
cat("Number of complete cases:", nrow(complete_data))
#complete case data: 7422 observations

# Simulation using complete case data

if(nrow(complete_data) == 7422) {
cat("The complete dataset contains 7422 observations. Ready for simulation!\n")
} else {
cat("The complete dataset has", nrow(complete_data), "observations. Check for missing values.\n")
}

library(writexl)
library(WriteXLS)
write_xlsx(complete_data, "Complete_Cases_PM25.xlsx")

Complete_Cases_PM25
cat("Number of complete cases:", nrow(complete_data))

generate_missing_pm25 <- function(data, missing_percent) {

set.seed(123) # Ensure reproducibility

# Total number of rows in the PM2.5 column

total_entries <- nrow(data)

# Calculate the number of missing values needed

num_missing <- round(total_entries * missing_percent / 100)

# Randomly select row indices for missing values

missing_rows <- sample(seq_len(total_entries), num_missing)

# Introduce missing values only in the PM2.5 column

data$`PM2.5`[missing_rows] <- NA

return(data)
}

data_missing_5 <- generate_missing_pm25(complete_data, 5)

View(data_missing_5)
data_missing_10 <- generate_missing_pm25(complete_data, 10)
View(data_missing_10)
data_missing_15 <- generate_missing_pm25(complete_data, 15)
View(data_missing_15)
data_missing_20 <- generate_missing_pm25(complete_data, 20)
View(data_missing_20)

# 500 Simulated patterns

simulate_missing_patterns <- function(data, missing_percent, num_simulations) {
# Initialize a list to store simulated datasets
simulations <- list()

for (i in 1:num_simulations) {
# Generate missing data
simulated_data <- generate_missing_pm25(data, missing_percent)
# Store in the list
simulations[[i]] <- simulated_data
}

return(simulations)
}

num_simulations <- 500

simulations_5 <- simulate_missing_patterns(complete_data, 5, num_simulations)
simulations_10 <- simulate_missing_patterns(complete_data, 10, num_simulations)
simulations_15 <- simulate_missing_patterns(complete_data, 15, num_simulations)
simulations_20 <- simulate_missing_patterns(complete_data, 20, num_simulations)

# 16/12/24

# Mean Imputation
mean_imputed <- complete_data
mean_value <- mean(mean_imputed$`PM2.5`, na.rm = TRUE)
mean_imputed$`PM2.5`[is.na(mean_imputed$`PM2.5`)] <- mean_value
mean_value

# Median Imputation
median_imputed <- complete_data
median_value <- median(median_imputed$`PM2.5`, na.rm = TRUE)
median_imputed$`PM2.5`[is.na(median_imputed$`PM2.5`)] <- median_value
median_value

install.packages("VIM")
library(VIM)
# kNN Imputation with k=5
knn_imputed <- kNN(complete_data, variable = "PM2.5", k = 5)

# View results
summary(knn_imputed$`PM2.5`)

install.packages("mice")
library(mice)
# Perform Multiple Imputation
mice_imputed <- mice(complete_data, m = 5, method = "pmm", maxit = 10, seed = 123)

# Extract the completed dataset

completed_data <- complete(mice_imputed, 1)

# View results
summary(completed_data$`PM2.5`)

# Check if complete_data exists and is not empty

if (exists("complete_data") && nrow(complete_data) > 0) {

# Create a copy of the dataset

simulated_data <- complete_data

# Simulate Missing Values (10% Missing in PM2.5)

set.seed(123)
missing_indices <- sample(1:nrow(simulated_data), size = 0.1 * nrow(simulated_data))

# Introduce Missing Values in PM2.5

simulated_data$`PM2.5`[missing_indices] <- NA

# Apply Mean Imputation

mean_imputed <- simulated_data
mean_value <- mean(mean_imputed$`PM2.5`, na.rm = TRUE)
mean_imputed$`PM2.5`[is.na(mean_imputed$`PM2.5`)] <- mean_value

# Calculate MAE
MAE <- mean(abs(complete_data$`PM2.5` - mean_imputed$`PM2.5`), na.rm = TRUE)
cat("Mean Absolute Error (MAE):", MAE, "\n")

} else {
cat("Error: Dataset 'complete_data' is empty or not loaded properly.\n")
}

# Check if complete_data exists and is not empty

if (exists("complete_data") && nrow(complete_data) > 0) {

# Create a copy of the dataset

simulated_data <- complete_data

# Simulate Missing Values (5% Missing in PM2.5)

 set.seed(123)
missing_indices <- sample(1:nrow(simulated_data), size = 0.05 * nrow(simulated_data))

# Introduce Missing Values in PM2.5

simulated_data$`PM2.5`[missing_indices] <- NA

# Apply Mean Imputation

mean_imputed <- simulated_data
mean_value <- mean(mean_imputed$`PM2.5`, na.rm = TRUE)
mean_imputed$`PM2.5`[is.na(mean_imputed$`PM2.5`)] <- mean_value

# Calculate MAE
MAE <- mean(abs(complete_data$`PM2.5` - mean_imputed$`PM2.5`), na.rm = TRUE)
cat("Mean Absolute Error (MAE):", MAE, "\n")

} else {
cat("Error: Dataset 'complete_data' is empty or not loaded properly.\n")
}
# Check if complete_data exists and is not empty
if (exists("complete_data") && nrow(complete_data) > 0) {

# Create a copy of the dataset

simulated_data <- complete_data

# Simulate Missing Values (15% Missing in PM2.5)

set.seed(123)
missing_indices <- sample(1:nrow(simulated_data), size = 0.15 * nrow(simulated_data))

# Introduce Missing Values in PM2.5

simulated_data$`PM2.5`[missing_indices] <- NA

# Apply Mean Imputation

mean_imputed <- simulated_data
mean_value <- mean(mean_imputed$`PM2.5`, na.rm = TRUE)
mean_imputed$`PM2.5`[is.na(mean_imputed$`PM2.5`)] <- mean_value

# Calculate MAE
MAE <- mean(abs(complete_data$`PM2.5` - mean_imputed$`PM2.5`), na.rm = TRUE)
cat("Mean Absolute Error (MAE):", MAE, "\n")

} else {
cat("Error: Dataset 'complete_data' is empty or not loaded properly.\n")
}
# Check if complete_data exists and is not empty
if (exists("complete_data") && nrow(complete_data) > 0) {

# Create a copy of the dataset

simulated_data <- complete_data

# Simulate Missing Values (20% Missing in PM2.5)

set.seed(123)
missing_indices <- sample(1:nrow(simulated_data), size = 0.2 * nrow(simulated_data))
 # Introduce Missing Values in PM2.5
simulated_data$`PM2.5`[missing_indices] <- NA

# Apply Mean Imputation

mean_imputed <- simulated_data
mean_value <- mean(mean_imputed$`PM2.5`, na.rm = TRUE)
mean_imputed$`PM2.5`[is.na(mean_imputed$`PM2.5`)] <- mean_value

# Calculate MAE
MAE <- mean(abs(complete_data$`PM2.5` - mean_imputed$`PM2.5`), na.rm = TRUE)
cat("Mean Absolute Error (MAE):", MAE, "\n")

} else {
cat("Error: Dataset 'complete_data' is empty or not loaded properly.\n")
}

# 17/12/24

original_data <- Worli_Mumbai_PM2_5

complete_data <- na.omit(original_data)
missing_percent <- colSums(is.na(original_data)) / nrow(original_data) * 100
missing_percent
cat("Number of complete cases:", nrow(complete_data))

# Verify the number of observations

if(nrow(complete_data) == 7422) {
cat("The complete dataset contains 7422 observations. Ready for simulation!\n")
} else {
cat("The complete dataset has", nrow(complete_data), "observations. Check for missing values.\n")
}

simulate_missing <- function(complete_data, percent) {

set.seed(123) # For reproducibility

# Calculate Number of Missing Values

n_missing <- floor(percent * nrow(complete_data))

# Generate Random Missing Indices

missing_indices <- sample(1:nrow(complete_data), n_missing)

# Create Simulated Data

simulated_data <- complete_data
simulated_data$`PM2.5`[missing_indices] <- NA

return(simulated_data)
}
# Function for Median Imputation
median_impute <- function(simulated_data) {
median_imputed <- simulated_data
 median_value <- median(median_imputed$`PM2.5`, na.rm = TRUE)
median_imputed$`PM2.5`[is.na(median_imputed$`PM2.5`)] <- median_value
return(median_imputed)
}
# Function to Calculate MAE
calculate_MAE <- function(original_data, imputed_data) {
MAE <- mean(abs(original_data$`PM2.5` - imputed_data$`PM2.5`), na.rm = TRUE)
return(MAE)
}

# Define Missing Percentages

missing_percents <- c(0.05, 0.10, 0.15, 0.20)
mae_results <- sapply(missing_percents, function(p) {
simulated_data <- simulate_missing(complete_data, p)
imputed_data <- median_impute(simulated_data)
calculate_MAE(complete_data, imputed_data)
})

# Print Results
for (i in 1:length(missing_percents)) {
cat("MAE for", missing_percents[i] * 100, "% Missing Data:", mae_results[i], "\n")
}

# Function to calculate RMSE

calculate_RMSE <- function(original_data, imputed_data) {
rmse <- sqrt(mean((original_data$`PM2.5` - imputed_data$`PM2.5`)^2, na.rm = TRUE))
return(rmse)
}

# Define missing percentages

missing_percents <- c(0.05, 0.10, 0.15, 0.20)

# Store RMSE results

rmse_results_median <- sapply(missing_percents, function(p) {
# Simulate missing data
simulated_data <- simulate_missing(complete_data, p)

# Perform median imputation

imputed_data <- median_impute(simulated_data)

# Calculate RMSE
rmse <- calculate_RMSE(complete_data, imputed_data)
return(rmse)
})

# Print RMSE Results

for (i in 1:length(missing_percents)) {
cat("RMSE for", missing_percents[i] * 100, "% Missing Data (Median Imputation):",
rmse_results_median[i], "\n")
}
# Function to calculate R^2
calculate_R2 <- function(original_data, imputed_data) {
ss_total <- sum((original_data$`PM2.5` - mean(original_data$`PM2.5`))^2, na.rm = TRUE)
ss_residual <- sum((original_data$`PM2.5` - imputed_data$`PM2.5`)^2, na.rm = TRUE)
R2 <- 1 - (ss_residual / ss_total)
return(R2)
}

# Define missing percentages

missing_percents <- c(0.05, 0.10, 0.15, 0.20)

# Store R^2 results

R2_results_median <- sapply(missing_percents, function(p) {
# Simulate missing data
simulated_data <- simulate_missing(complete_data, p)

# Perform median imputation

imputed_data <- median_impute(simulated_data)

# Calculate R^2
R2 <- calculate_R2(complete_data, imputed_data)
return(R2)
})

# Print R^2 Results

for (i in 1:length(missing_percents)) {
cat("R^2 for", missing_percents[i] * 100, "% Missing Data (Median Imputation):", R2_results_median[i],
"\n")
}

# Function to calculate Agreement Index

calculate_agreement_index <- function(original_data, imputed_data) {
y <- original_data$`PM2.5`
y_hat <- imputed_data$`PM2.5`
y_mean <- mean(y, na.rm = TRUE)

numerator <- sum((y - y_hat)^2, na.rm = TRUE)

denominator <- sum((abs(y - y_mean) + abs(y_hat - y_mean))^2, na.rm = TRUE)

d <- 1 - (numerator / denominator)

return(d)
}

# Define missing percentages

missing_percents <- c(0.05, 0.10, 0.15, 0.20)

# Store Agreement Index results

agreement_results_median <- sapply(missing_percents, function(p) {
# Simulate missing data
simulated_data <- simulate_missing(complete_data, p)
 # Perform median imputation
imputed_data <- median_impute(simulated_data)

# Calculate Agreement Index

d <- calculate_agreement_index(complete_data, imputed_data)
return(d)
})

# Print Agreement Index Results

for (i in 1:length(missing_percents)) {
cat("Agreement Index for", missing_percents[i] * 100, "% Missing Data (Median Imputation):",
agreement_results_median[i], "\n")
}

# Function for Mean Imputation

mean_impute <- function(simulated_data) {
mean_imputed <- simulated_data
mean_value <- mean(mean_imputed$`PM2.5`, na.rm = TRUE)
mean_imputed$`PM2.5`[is.na(mean_imputed$`PM2.5`)] <- mean_value
return(mean_imputed)
}
# Function to Calculate MAE
calculate_MAE <- function(original_data, imputed_data) {
MAE <- mean(abs(original_data$`PM2.5` - imputed_data$`PM2.5`), na.rm = TRUE)
return(MAE)
}

# Define Missing Percentages

missing_percents <- c(0.05, 0.10, 0.15, 0.20)
mae_results <- sapply(missing_percents, function(p) {
simulated_data <- simulate_missing(complete_data, p)
imputed_data <- mean_impute(simulated_data)
calculate_MAE(complete_data, imputed_data)
})

# Print Results
for (i in 1:length(missing_percents)) {
cat("MAE for", missing_percents[i] * 100, "% Missing Data:", mae_results[i], "\n")
}

# Function to calculate RMSE

calculate_RMSE <- function(original_data, imputed_data) {
rmse <- sqrt(mean((original_data$`PM2.5` - imputed_data$`PM2.5`)^2, na.rm = TRUE))
return(rmse)
}

# Missing percentages
missing_percents <- c(0.05, 0.10, 0.15, 0.20)

# Store RMSE results

rmse_results <- sapply(missing_percents, function(p) {
 # Simulate missing data
simulated_data <- simulate_missing(complete_data, p)

# Perform mean imputation

imputed_data <- mean_impute(simulated_data)

# Calculate RMSE
rmse <- calculate_RMSE(complete_data, imputed_data)
return(rmse)
})

# Print RMSE Results

for (i in 1:length(missing_percents)) {
cat("RMSE for", missing_percents[i] * 100, "% Missing Data:", rmse_results[i], "\n")
}

# Function to calculate R^2

calculate_R2 <- function(original_data, imputed_data) {
ss_total <- sum((original_data$`PM2.5` - mean(original_data$`PM2.5`))^2, na.rm = TRUE)
ss_residual <- sum((original_data$`PM2.5` - imputed_data$`PM2.5`)^2, na.rm = TRUE)
R2 <- 1 - (ss_residual / ss_total)
return(R2)
}

# Define missing percentages

missing_percents <- c(0.05, 0.10, 0.15, 0.20)

# Store R^2 results

R2_results_mean <- sapply(missing_percents, function(p) {
# Simulate missing data
simulated_data <- simulate_missing(complete_data, p)

# Perform mean imputation

imputed_data <- mean_impute(simulated_data)

# Calculate R^2
R2 <- calculate_R2(complete_data, imputed_data)
return(R2)
})

# Print R^2 Results

for (i in 1:length(missing_percents)) {
cat("R^2 for", missing_percents[i] * 100, "% Missing Data (Mean Imputation):", R2_results_mean[i], "\n")
}

# Function to calculate Agreement Index

calculate_agreement_index <- function(original_data, imputed_data) {
y <- original_data$`PM2.5`
y_hat <- imputed_data$`PM2.5`
y_mean <- mean(y, na.rm = TRUE)
 numerator <- sum((y - y_hat)^2, na.rm = TRUE)
denominator <- sum((abs(y - y_mean) + abs(y_hat - y_mean))^2, na.rm = TRUE)

d <- 1 - (numerator / denominator)

return(d)
}

# Define missing percentages

missing_percents <- c(0.05, 0.10, 0.15, 0.20)

# Store Agreement Index results

agreement_results <- sapply(missing_percents, function(p) {
# Simulate missing data
simulated_data <- simulate_missing(complete_data, p)

# Perform mean imputation

imputed_data <- mean_impute(simulated_data)

# Calculate Agreement Index

d <- calculate_agreement_index(complete_data, imputed_data)
return(d)
})

# Print Agreement Index Results

for (i in 1:length(missing_percents)) {
cat("Agreement Index for", missing_percents[i] * 100, "% Missing Data (Mean Imputation):",
agreement_results[i], "\n")
}

NOT SURE

# Function for kNN Imputation

knn_impute <- function(simulated_data, k = 5) {
imputed_data <- knnImputation(simulated_data, k = k)
return(imputed_data)
}
# Function to Calculate MAE
calculate_MAE <- function(original_data, imputed_data) {
MAE <- mean(abs(original_data$`PM2.5` - imputed_data$`PM2.5`), na.rm = TRUE)
return(MAE)
}

install.packages("DMwR2")
library(DMwR2)

# Perform kNN Imputation

knn_impute <- function(data, k) {
imputed_data <- knnImputation(data, k = k)
return(imputed_data)
}

# Install and Load VIM

install.packages("VIM")
library(VIM)

# Load required package

library(DMwR2)
set.seed(123) # For reproducibility

# Convert to a dataframe if not already

complete_data <- as.data.frame(complete_data)

# Obtain only complete cases

complete_data <- na.omit(complete_data)

# Introduce 20% missing values randomly

missing_data <- complete_data
n <- nrow(missing_data) * ncol(missing_data)
num_missing <- round(0.20 * n)

# Randomly select indices to introduce NAs

indices <- sample(1:n, num_missing, replace = FALSE)

# Set the selected indices to NA

missing_data[indices] <- NA

# Check for missing values

sum(is.na(missing_data)) # Should be 20% of total elements
# Perform kNN Imputation
imputed_data <- kNN(simulated_data, variable = "PM2.5", k = 5)

# Define Missing Percentages

missing_percents <- c(0.05, 0.10, 0.15)
k_value <- 5 # You can change this to try different values

mae_results <- sapply(missing_percents, function(p) {

simulated_data <- simulate_missing(complete_data, p)
imputed_data <- knn_impute(simulated_data, k = 5)
calculate_MAE(complete_data, imputed_data)
})

# Print Results
for (i in 1:length(missing_percents)) {
cat("MAE for", missing_percents[i] * 100, "% Missing Data with k =", k_value, ":", mae_results[i], "\n")
}