mccf1

mccf1: a method to evaluate the performance of binary classification models

Summary

The mccf1 reads an input dataset of real values and prediction values of binary classification, and outputs the corresponding MCC-F1 score metric and the best confusion matrix threshold of the prediction. It can also plot the MCC-F1 score curve.

Installation

The mccf1 package is now available on CRAN: https://cran.r-project.org/package=mccf1

To run mccf1, you need to have the following programs and packages installed in your machine:

• R (version 3.3.3)

• R ROCR package

    install.packages("ROCR")
  

• R ggplot2 package

    install.packages("ggplot2")
  

After installing R, you can install the mccf1 package by typing in the R terminal: install.packages("mccf1")

Execution instructions

Allocate your ground-truth real values into an array, and your predicted real values into another one. Then call the mccf1_calcu() function to compute the MCC-F1 score metric and the best thresold, or the mccf1_plot() to plot and save the MCC-F1 score curve. See the example below.

An example

To run mccf1, you first need to have a vector of actual values and a vector of predicted values.

We use beta distribution to simulate the predicted value vector. From the R terminal, type:

positive_class <- 1
negative_class <- 0
num_of_positive_class <- 1000
num_of_negative_class <- 10000
proportion_of_predicted_for_pos_type_1 <- 0.3
proportion_of_predicted_for_pos_type_2 <- 1 - proportion_of_predicted_for_pos_type_1
shape1_pos_type_1 <- 12
shape2_pos_type_1 <- 2
shape1_pos_type_2 <- 3
shape2_pos_type_2 <- 4
shape1_neg <- 2 
shape2_neg <- 3
actual <- c(rep(positive_class, num_of_positive_class), rep(negative_class, num_of_negative_class))
predicted <- c(rbeta(proportion_of_predicted_for_pos_type_1 * num_of_positive_class, shape1_pos_type_1,    
              shape2_pos_type_1), rbeta(proportion_of_predicted_for_pos_type_2 * num_of_positive_class, 
              shape1_pos_type_2, shape2_pos_type_2), rbeta(num_of_negative_class, shape1_neg, shape2_neg))

Secondly, include the code file mccf1.R.

source("mccf1.R")

Then you can use the function mccf1_calcu() to generate the corresponding MCC-F1 score metric and the best threshold of the prediction:

result <- mccf1_calcu(actual, predicted)

cat("The MCC F1 score metric is ", result$mccf1_metric," \n");

cat("The best confusion matrix threshold for the MCC F1 score curve is ", result$bestThreshold," \n");

You can also change the fold when calculating MCC-F1 metric.

mccf1_calcu(actual, predicted, fold = 50)

You can also use mccf1_plot() to plot the corresponding MCC-F1 curve.

mccf1_plot(actual, predicted, .title="the MCC-F1 score curve (example)")

You can save the pdf file of the image to a specified folder.

pdfFileName = "./curve.pdf" 

mccf1_plot(actual, predicted, .title="the MCC-F1 score curve (example)", .curveFileName=pdfFileName)

Contacts

The mccf1 package was developed by Chang Cao, Davide Chicco, and Michael M. Hoffman at the Hoffman Lab of the Princess Margaret Cancer Centre (Toronto, Ontario, Canada). Questions should be addressed to Michael M. Hoffman [email protected].