Stats_Project.Rmd

```{r}
# import dataset in to R
library(readr)
kidney_disease <- read_csv("C:/Users/eshan/Downloads/kidney_disease.csv")
```


```{r}
# view dataset after importing dataset
View(kidney_disease)
```

```{r}
# dimension of the dataset
dim(kidney_disease)

# number of rows and columns
nrow(kidney_disease)
ncol(kidney_disease)
```
```{r}
# overview of the variables
str(kidney_disease)
```
```{r}
# First 6 rows of the dataset
head(kidney_disease)
```
```{r}
#summary of the dataset
summary(kidney_disease)
```

```{r}
# total count of null values
sum(is.na(kidney_disease))
```
```{r}
# null values per column
colSums(is.na(kidney_disease))
```

```{r}
library(readr)
kidney_disease_cleaned <- read_csv("C:/Users/eshan/Downloads/kidney_disease_cleaned.csv")
View(kidney_disease_cleaned)
```

```{r}
# overview of the variables
str(kidney_disease_cleaned)
```
```{r}
#changing data type
library(dplyr)

# Assuming 'df' is your data frame
kidney_disease_cleaned <- kidney_disease_cleaned %>%
  mutate(across(c(id, age, bp, sg, al, su, bgr, bu, sc, sod, pot, hemo), as.numeric),
         across(c(rbc, pc, pcc, ba, htn, dm, cad, appet, pe, ane, classification), as.factor))

str(kidney_disease_cleaned)
```
```{r}
#summary of the dataset
summary(kidney_disease_cleaned)
```


```{r}
# checking for outlier detection
# Filter numeric columns
numeric_data <- kidney_disease_cleaned[sapply(kidney_disease_cleaned, is.numeric)]

# Check for outliers in numeric columns and get outlier indices
outlier_counts <- sapply(numeric_data, function(x) {
  iqr <- IQR(x)
  lower_bound <- quantile(x, 0.25) - 1.5 * iqr
  upper_bound <- quantile(x, 0.75) + 1.5 * iqr
  outliers <- x < lower_bound | x > upper_bound
  outlier_count <- sum(outliers, na.rm = TRUE)
  return(outlier_count)
})

# Print outlier count for each numeric column
print("Outlier count for each numeric column:")
print(outlier_counts)
print(sum(outlier_counts))
```
```{r}
#data visualization for outliers
# Select specific columns for box plots
columns_to_plot <- c("age", "bp", "sg", "al", "su", "bgr", "bu", "sc", "sod", "pot", "hemo", "pcv", "wc", "rc")

# Create a list to store the box plot data
box_plot_data <- lapply(columns_to_plot, function(col) kidney_disease_cleaned[[col]])
names(box_plot_data) <- columns_to_plot


# Create box plots
for (col in columns_to_plot) {
  boxplot(kidney_disease_cleaned[[col]], main = col, ylab = "Value")
}

```

```{r}
# Checking for outlier detection with winsorization
library(dplyr)

# Function to perform winsorization on a vector
winsorize <- function(x, p_low = 0.05, p_high = 0.95) {
  lower_bound <- quantile(x, p_low)
  upper_bound <- quantile(x, p_high)
  x[x < lower_bound] <- lower_bound
  x[x > upper_bound] <- upper_bound
  return(x)
}

# Apply winsorization to numeric columns
winsorized_data <- numeric_data %>% 
  mutate_all(winsorize)

# Check for outliers in winsorized data and get outlier counts
outlier_counts_winsorized <- sapply(winsorized_data, function(x) {
  outliers <- x < quantile(x, 0.05) | x > quantile(x, 0.95)
  outlier_count <- sum(outliers, na.rm = TRUE)
  return(outlier_count)
})

# Print outlier count for each numeric column after winsorization
print("Outlier count for each numeric column after winsorization:")
print(outlier_counts_winsorized)
print(sum(outlier_counts_winsorized))

```



```{r}
# identification of outliers, replacing outliers with upper and lower bound

# Function to perform winsorization on a vector
winsorize_replace <- function(x, p_low = 0.05, p_high = 0.95) {
  lower_bound <- quantile(x, p_low)
  upper_bound <- quantile(x, p_high)
  x[x < lower_bound] <- lower_bound
  x[x > upper_bound] <- upper_bound
  return(x)
}

# Apply winsorization and replacement to all numeric columns
numeric_columns <- sapply(kidney_disease_cleaned, is.numeric)
kidney_disease_cleaned[, numeric_columns] <- lapply(kidney_disease_cleaned[, numeric_columns], winsorize_replace)

# Summary of the dataset after replacing outliers
summary(kidney_disease_cleaned)

```

```{r}

#data visualization for outliers

# Select specific columns for box plots
columns_to_plot <- c("age", "bp", "sg", "al", "su", "bgr", "bu", "sc", "sod", "pot", "hemo", "pcv", "wc", "rc")

# Create a list to store the box plot data
box_plot_data <- lapply(columns_to_plot, function(col) kidney_disease_cleaned[[col]])
names(box_plot_data) <- columns_to_plot


# Create box plots
for (col in columns_to_plot) {
  boxplot(kidney_disease_cleaned[[col]], main = col, ylab = "Value")
}
```


```{r}
numeric_columns <- c("id", "age", "bp", "sg", "al", "su", "bgr", "bu", "sc", "sod", "pot", "hemo")

# Create scatter plots in a loop
for (column in numeric_columns) {
  # Create scatter plot
  plot(kidney_disease_cleaned[[column]], 
       main = paste("Scatter Plot of", column), 
       xlab = "Index", 
       ylab = column)
  
  # Add grid
  grid()
}

```



```{r}
# Load necessary library
library(ggplot2)

# Plot histograms for numeric variables
numeric_data <- subset(kidney_disease_cleaned, select = sapply(kidney_disease_cleaned, is.numeric))
numeric_data_long <- tidyr::gather(numeric_data, key = "variable", value = "value")

ggplot(numeric_data_long, aes(x = value)) +
  geom_histogram(bins = 20, fill = "pink", color = "black") +
  facet_wrap(~ variable, scales = "free") +
  labs(title = "Histograms of Numeric Variables")
```
```{r}
# Plot barplot for categorical variables
# Identify categorical columns
categorical_columns <- sapply(kidney_disease_cleaned, is.factor)

# Extract categorical data
categorical_data <- kidney_disease_cleaned[, categorical_columns]

# Set margins to adjust plot size
par(mar = c(5, 5, 2, 2))  # Set margins (bottom, left, top, right)

# Plot bar plots for each categorical variable
invisible(
  lapply(names(categorical_data), function(col) {
    barplot(table(categorical_data[[col]]), main = col, xlab = col, ylab = "Frequency", col = "skyblue", border = "darkblue", ylim = c(0, max(table(categorical_data[[col]])) + 5))
  })
)

# Reset the layout to default
par(mar = c(5, 4, 4, 2) + 0.1)  # Default margins

```


```{r}
# corelation analysis between numerical variables and outcome

# Load required libraries
library(ggplot2)

# Assuming 'kidney_disease_cleaned' is your dataset containing numerical variables and the 'classification' outcome

# Select numerical variables
numerical_vars <- kidney_disease_cleaned[, sapply(kidney_disease_cleaned, is.numeric)]

# Remove variables with zero standard deviation
numerical_vars <- numerical_vars[, apply(numerical_vars, 2, sd) != 0]

# Calculate correlation between numerical variables and classification
correlation_with_classification <- cor(numerical_vars, as.numeric(kidney_disease_cleaned$classification))

# Print correlation result
print(correlation_with_classification)

```

```{r}
# Load required libraries
library(ggplot2)

# Assuming 'kidney_disease_cleaned' is your dataset containing numeric variables
# Select numeric variables
numeric_columns <- sapply(kidney_disease_cleaned, is.numeric)
numeric_data <- kidney_disease_cleaned[, numeric_columns]
numeric_data_target <- cbind(numeric_data, classification = as.numeric(kidney_disease_cleaned$classification))

# Compute the correlation matrix
correlation_matrix <- cor(numeric_data_with_target)

# Convert the correlation matrix to a long format
melted_corr <- as.data.frame(as.table(correlation_matrix))
colnames(melted_corr) <- c("Var1", "Var2", "value")

# Plot the heatmap
ggplot(data = melted_corr, aes(x = Var1, y = Var2, fill = value)) +
 geom_tile() +
  geom_text(aes(label = round(value, 2)), color = "black", size = 3) +  # Add correlation values
  scale_fill_gradient2(low = "blue", high = "red", mid = "white", midpoint = 0, limits = c(-1, 1), name = "Correlation") +
  theme_minimal() +
  labs(title = "Correlation Heatmap of Numeric Variables",
       x = "", y = "") +
  theme(axis.text.x = element_text(angle = 45, vjust = 1, hjust = 1))

```


```{r}
# List of categorical variables (replace with your actual variable names)
categorical_vars <- c("htn", "dm", "cad", "appet", "pe", "ane", "su", "rbc", "pc","pcc", "ba")

# Loop through each categorical variable
for (var in categorical_vars) {
  # Create contingency table
  contingency_table <- table(kidney_disease_cleaned[[var]], kidney_disease_cleaned$classification)
  
  # Perform chi-square test
  chi_square_result <- chisq.test(contingency_table)
  
  # Print variable name and chi-square test result
  cat("Chi-square test for", var, "vs. CKD status:\n")
  print(chi_square_result)
  cat("\n")
}
```

```{r}
# data visualization for categorical variable with classification

# Load required libraries
library(ggplot2)

# List of categorical variables (replace with your actual variable names)
categorical_vars <- c("htn", "dm", "cad", "appet", "pe", "ane", "rbc", "pc", "pcc", "ba")

# Create an empty data frame to store the results
chi_square_results <- data.frame(Variable = character(), ChiSquareStatistic = numeric(), p_value = numeric())

# Loop through each categorical variable
for (var in categorical_vars) {
  # Perform chi-square test
  chi_square_result <- chisq.test(table(kidney_disease_cleaned[[var]], kidney_disease_cleaned$classification))
  
  # Store variable name, chi-square test statistic, and p-value in the data frame
  chi_square_results <- rbind(chi_square_results, data.frame(Variable = var, ChiSquareStatistic = chi_square_result$statistic, p_value = chi_square_result$p.value))
}

# Order the data frame by ChiSquareStatistic in descending order
chi_square_results <- chi_square_results[order(-chi_square_results$ChiSquareStatistic), ]

# Plot bar plot with color gradient
ggplot(chi_square_results, aes(x = reorder(Variable, -ChiSquareStatistic), y = ChiSquareStatistic, fill = ChiSquareStatistic)) +
  geom_bar(stat = "identity") +
  scale_fill_gradient(low = "lightblue", high = "darkblue") +
  labs(x = "Categorical Variable", y = "Chi-Square Test Statistic", title = "Association with Classification Outcome") +
  theme(axis.text.x = element_text(angle = 90, vjust = 0.5, hjust = 1))

```


```{r}
# normality testing

# Load necessary library
library(dplyr)

# Filter out variables with no variability
numeric_data_filtered <- select_if(numeric_data, function(x) length(unique(x)) > 1)

# Perform Shapiro-Wilk test for each numerical variable
shapiro_results <- lapply(numeric_data_filtered, shapiro.test)

# Combine results into a data frame
shapiro_df <- bind_rows(lapply(names(shapiro_results), function(var) {
  data.frame(variable = var,
             statistic = shapiro_results[[var]]$statistic,
             p.value = shapiro_results[[var]]$p.value,
             stringsAsFactors = FALSE)
}))

# Print results
print(shapiro_df)
```


```{r}
#data visualization after log transformation
library(ggplot2)
library(tidyr)

# Select only numeric variables
numeric_data <- kidney_disease_cleaned %>%
  select_if(is.numeric)

# Log transformation for numeric variables
numeric_data_log <- log(numeric_data + 1)  # Adding 1 to avoid log(0)

# Prepare data for plotting
log_data_long <- tidyr::gather(numeric_data_log, key = "variable", value = "value")

# Plot histograms of log-transformed numeric variables
ggplot(log_data_long, aes(x = value)) +
  geom_histogram(bins = 20, fill = "pink", color = "black") +
  facet_wrap(~ variable, scales = "free") +
  labs(title = "Histograms of Log-Transformed Numeric Variables")

```

## Research  Question 1:What is the association between blood glucose levels (bgr) and blood urea levels (bu) with chronic kidney disease?

Hypothesis Testing:

Null Hypothesis (H0): There is no significant association between blood glucose levels (bgr) and blood urea levels (bu) with chronic kidney disease.

Alternative Hypothesis (H1): There is a significant association between blood glucose levels (bgr) and blood urea levels (bu) with chronic kidney disease.


```{r}
correlation_coefficient <- cor(kidney_disease_cleaned$bgr, kidney_disease_cleaned$bu, method = "spearman")

# Print the correlation coefficient
print(correlation_coefficient)

```
Analysis of correlation: Spearman correlation measures the strength and direction of association between two ranked variables, a value of 0.2236 indicates a positive association, meaning that higher values of bgr tend to be associated with higher values of bu, and vice versa. However, the correlation is considered weak, suggesting that the relationship between these variables is not very strong.

```{r}
# Load necessary library
library(ggplot2)

# Scatter plot for bu and bgr
ggplot(kidney_disease_cleaned, aes(x = bu, y = bgr)) +
  geom_point() +
  labs(x = "Blood Urea", y = "Blood Glucose Random", title = "Scatter Plot of Blood Urea vs Blood Glucose Random")

```

```{r}
# Fit logistic regression model
logistic_model <- glm(classification ~ bgr + bu, data = kidney_disease_cleaned, family = binomial(link = "logit"))

# Summarize the model
summary(logistic_model)

library(car)
vif_values <- vif(logistic_model)
vif_values
```
Analysis of Logistic regression:

The logistic regression model reveals that for each unit increase in blood glucose levels (bgr), the log odds of chronic kidney disease increase by approximately 0.038 (p < 0.001). Similarly, for each unit increase in blood urea levels (bu), the log odds of chronic kidney disease increase by approximately 0.053 (p < 0.001). The intercept of -6.357 (p < 0.001) represents the log odds of chronic kidney disease when both bgr and bu are zero. The model's null deviance is 529.25 on 399 degrees of freedom, and the residual deviance is 374.64 on 397 degrees of freedom, with a dispersion parameter of 1. AIC for the model is 380.64.



## Research Question 2: What is the association between bacteria (ba) and hypertension (htn) with chronic kidney disease?


Null Hypothesis (H0): There is no significant association between ba and htn with chronic kidney disease.

Alternative Hypothesis (H1): There is a significant association between ba and htn with chronic kidney disease.



```{r}
# Chi square test for checking independence
chisq.test(kidney_disease_cleaned$htn, kidney_disease_cleaned$ba)
```


```{r}
# Fit logistic regression model
logistic_model <- glm(classification ~ htn+ba, data = kidney_disease_cleaned, family = binomial(link = "logit"))

# Summarize the model
summary(logistic_model)
```

Analysis of Logistic Regression:

In the logistic regression model fitted to predict the classification outcome, the intercept is estimated to be -39.12. This implies that when all predictor variables, namely hypertension (htn) and blood albumin (ba), are zero, the log-odds of the outcome is expected to be approximately -39.12. However, both the htn and ba variables exhibit coefficients with p-values greater than 0.05 (0.982 and 0.992, respectively), indicating that they are not statistically significant predictors of the outcome. Specifically, for every one-unit increase in htn, the log-odds of the outcome are expected to increase by 19.90, while for every one-unit increase in ba, the log-odds of the outcome are expected to increase by 18.74. However, due to their high p-values, these relationships are not deemed statistically significant. Therefore, based on this model, neither hypertension nor blood albumin is considered a significant predictor of the classification outcome.