Mining NB-Frequent Itemsets and NB-Precise Rules

NBMiner is an implementation of the model-based mining algorithm for mining NB-frequent itemsets and NB-precise rules. Michael Hahsler (2006) .

R package arulesNBMiner - Mining NB-Frequent Itemsets and NB-Precise Rules

License

Maintainer: Michael Hahsler

This R package extends arules with NBMiner, an implementation of the model-based mining algorithm for NB-frequent itemsets described in Michael Hahsler’s paper, “A model-based frequency constraint for mining associations from transaction data.” Data Mining and Knowledge Discovery, 13(2):137-166, September 2006.

This algorithm dynamically chooses support for itemsets based on the deviation from a fitted independence model. NBMiner is better at suppressing spurious patters and finds interesting patterns with lower support compared to the traditional minimum support-based algorithms.

Installation

Stable CRAN version: Install from within R with

install.packages("arulesNBMiner")

Current development version: Install from r-universe.

install.packages("arulesNBMiner",
    repos = c("https://mhahsler.r-universe.dev",
              "https://cloud.r-project.org/"))

Usage

Estimate negative binomial distribution (NBD) model parameters for the arules Groceries data set.

library("arulesNBMiner")
data("Groceries")

param <- NBMinerParameters(Groceries, trim = 0.1)

## 17 item(s) trimmed, leaving  152  items. 
## using Expectation Maximization for missing zero class
## iteration = 1 , zero class = 2 , k = 0.86 , m = 153 
## total items =  154 
## 
## Goodness of fit/chi-square test on binned count data
##   H0: Observed counts match the expected proportions.
##   Bins:  10 
##   X-squared =  6.6  with  9  degrees of freedom
##   p.value =  0.68

Note that the model shows independence but the dataset contains a significant amount of correlated items shown by the larger cumulative frequency.

Mine NB-frequent itemsets

itemsets_NB <- NBMiner(Groceries, parameter = param, minlen = 2L)
itemsets_NB

## set of 2881 itemsets

Add support and inspect some NB-frequent itemsets with the highest precision.

quality(itemsets_NB) <- cbind(quality(itemsets_NB), support = interestMeasure(itemsets_NB,
    "support", transactions = Groceries))

inspect(head(itemsets_NB, by = "precision"))

##     items                                        precision support
## [1] {bottled beer, liquor, red/blush wine}       1         0.00193
## [2] {bottled beer, liquor}                       1         0.00468
## [3] {soda, bottled beer, liquor, red/blush wine} 1         0.00081
## [4] {whole milk, curd}                           1         0.02613
## [5] {whole milk, rolls/buns, margarine}          1         0.00793
## [6] {whole milk, butter}                         1         0.02755

A precision close to 1 means that there is very little chance that they are spurious. We see that some very low support itemsets are under the top NB-frequent itemsets.