jamovi

Code: Select all

MahalanobisOutlierDetector <- R6::R6Class("MahalanobisOutlierDetector",
  public = list(
    data = NULL,
    threshold = 5.0,
    mDist = NULL,  # Abbreviated name for Mahalanobis distances
    outliers = NULL,

    initialize = function(data, threshold = 5.0) {
      self$data <- data
      self$threshold <- threshold
      self$calculate_distances()
      self$identify_outliers()
    },

    calculate_distances = function() {
      # Calculate the Mahalanobis distances on the centered data
      self$mDist <- stats::mahalanobis(self$data,
                                       colMeans(self$data), stats::cov(self$data))

      # Add the Mahalanobis distances to the dataset
      self$data$mDist <- self$mDist
    },

    identify_outliers = function() {
      # Identify outliers based on the threshold
      self$data$outliers <- ifelse(self$data$mDist > self$threshold, 
                                   "True", 
                                   "False")
      self$outliers <- self$data[self$data$outliers == "True", ]
    },
    
    plot_qq = function() {
      ggplot2::ggplot(self$data, ggplot2::aes(x = mDist)) +
        ggplot2::geom_histogram(bins = 30, fill = 'blue', alpha = 0.7) +
        ggplot2::geom_vline(xintercept = self$threshold, 
                           color = 'red', 
                           linetype = 'dashed') +
        ggplot2::labs(title = 'Mahalanobis Distance Histogram',
                      x = 'Mahalanobis Distance',
                      y = 'Frequency') +
        ggplot2::theme_minimal()
    }
  )
)

# ---------------------------------------------------------
# Example usage with a dataset of 150 cases and 4 variables
set.seed(123) # For reproducibility
# Create an instance of the MahalanobisOutlierDetector class
detector <- MahalanobisOutlierDetector$new(data, 
                                           threshold = 5.0)
# Print only the outliers (rows where outliers == "True")
print(detector$outliers)

# Plot the Mahalanobis distance histogram with a threshold line
detector$plot_qq()

data <- cbind(detector$data)