Explainable Outlier Detection in Titanic dataset

Explainable Outlier Detection in Titanic dataset

This short notebook illustrates basic usage of the OutlierTree library for explainable outlier detection using the Titanic dataset. For more details, you can check the package’s documentation at CRAN or through R’s help (e.g. ?outliertree::outlier.tree). For a more interesting and interactive example, see the documentation of the main function (outlier.tree), which uses a larger dataset.

The dataset is very popular and can be downloaded from different sources, such as Kaggle or many university webpages. This vignette took it from the following link: https://github.com/jbryer/CompStats/raw/master/Data/titanic3.csv

The data comes bundled in the package so there is no need to download it from the link above.

Loading the raw data

library(data.table)
library(kableExtra)
library(outliertree)
data("titanic")

titanic |>
    head(5) |>
    kable() |>
    kable_styling()

pclass	survived	name	sex	age	sibsp	parch	ticket	fare	cabin	embarked	boat	body	home.dest
1	1	Allen, Miss. Elisabeth Walton	female	29.00	0	0	24160	211.3375	B5	S	2	NA	St Louis, MO
1	1	Allison, Master. Hudson Trevor	male	0.92	1	2	113781	151.5500	C22 C26	S	11	NA	Montreal, PQ / Chesterville, ON
1	0	Allison, Miss. Helen Loraine	female	2.00	1	2	113781	151.5500	C22 C26	S	NA	NA	Montreal, PQ / Chesterville, ON
1	0	Allison, Mr. Hudson Joshua Creighton	male	30.00	1	2	113781	151.5500	C22 C26	S	NA	135	Montreal, PQ / Chesterville, ON
1	0	Allison, Mrs. Hudson J C (Bessie Waldo Daniels)	female	25.00	1	2	113781	151.5500	C22 C26	S	NA	NA	Montreal, PQ / Chesterville, ON

Pre-processing the data

## Capitalize column names and some values for easier reading
capitalize <- function(x) gsub("^(\\w)", "\\U\\1\\E", x, perl=TRUE)

titanic <- as.data.table(titanic)
titanic[
    , setnames(.SD, names(.SD), capitalize(names(.SD)))
][
    , setnames(.SD, "Sibsp", "SibSp")
][
    , Sex := capitalize(Sex)
] -> titanic

## Convert 'survived' to yes/no for easier reading
titanic[
    , Survived := ifelse(Survived, "Yes", "No")
]

## Some columns are not useful, such as name (an ID), ticket number (another ID),
## or destination (too many values, many non-repeated)
titanic[
    , !c("Name", "Ticket", "Home.dest")
] -> titanic

## Ordinal columns need to be passed as ordered factors
cols_ord <- c("Pclass", "Parch", "SibSp")
titanic[
    , (cols_ord) := lapply(.SD, function(x) factor(x, ordered = TRUE))
    , .SDcols = cols_ord
]

## A look at the processed data
titanic |>
    head(5) |>
    kable() |>
    kable_styling()

Pclass	Survived	Sex	Age	SibSp	Parch	Fare	Cabin	Embarked	Boat	Body
1	Yes	Female	29.00	0	0	211.3375	B5	S	2	NA
1	Yes	Male	0.92	1	2	151.5500	C22 C26	S	11	NA
1	No	Female	2.00	1	2	151.5500	C22 C26	S	NA	NA
1	No	Male	30.00	1	2	151.5500	C22 C26	S	NA	135
1	No	Female	25.00	1	2	151.5500	C22 C26	S	NA	NA

Fitting a model

library(outliertree)

## Fit model with default hyperparameters
otree <- outlier.tree(titanic)
otree

Reporting top 9 outliers [out of 9 found]

row [171] - suspicious column: [Fare] - suspicious value: [0.00]
    distribution: 98.571% >= 25.74 - [mean: 55.22] - [sd: 27.56] - [norm. obs: 69]
    given:
        [Pclass] = [1]
        [Boat] in [1, 15, 5, 5 7, 5 9, 7, 8 10, 9, B, C] (value: C)


row [19] - suspicious column: [Age] - suspicious value: [32.00]
    distribution: 96.000% >= 43.00 - [mean: 48.35] - [sd: 3.16] - [norm. obs: 24]
    given:
        [Cabin] in [A16, A20, B10, B52 B54 B56, B82 B84, C110, C116, C124, C126, C86, C92, D15, D17, D33, D46, E12, E31, E58, E63] (value: D15)


row [897] - suspicious column: [Fare] - suspicious value: [0.00]
    distribution: 99.216% >= 3.17 - [mean: 9.68] - [sd: 6.98] - [norm. obs: 506]
    given:
        [Pclass] = [3]
        [SibSp] = [0]


row [899] - suspicious column: [Fare] - suspicious value: [0.00]
    distribution: 99.216% >= 3.17 - [mean: 9.68] - [sd: 6.98] - [norm. obs: 506]
    given:
        [Pclass] = [3]
        [SibSp] = [0]


row [964] - suspicious column: [Fare] - suspicious value: [0.00]
    distribution: 99.216% >= 3.17 - [mean: 9.68] - [sd: 6.98] - [norm. obs: 506]
    given:
        [Pclass] = [3]
        [SibSp] = [0]


row [1255] - suspicious column: [Fare] - suspicious value: [0.00]
    distribution: 99.216% >= 3.17 - [mean: 9.68] - [sd: 6.98] - [norm. obs: 506]
    given:
        [Pclass] = [3]
        [SibSp] = [0]


row [1045] - suspicious column: [Fare] - suspicious value: [15.50]
    distribution: 96.774% <= 8.52 - [mean: 7.73] - [sd: 0.28] - [norm. obs: 30]
    given:
        [Pclass] = [3]
        [SibSp] = [0]
        [Boat] in [10, 13 15, 13 15 B, 15 16, 16, 6, 9, A, B] (value: 16)


row [1147] - suspicious column: [Fare] - suspicious value: [29.12]
    distribution: 97.849% <= 15.50 - [mean: 7.89] - [sd: 1.17] - [norm. obs: 91]
    given:
        [Pclass] = [3]
        [SibSp] = [0]
        [Embarked] = [Q]


row [1164] - suspicious column: [Fare] - suspicious value: [24.15]
    distribution: 97.849% <= 15.50 - [mean: 7.89] - [sd: 1.17] - [norm. obs: 91]
    given:
        [Pclass] = [3]
        [SibSp] = [0]
        [Embarked] = [Q]

Outlier Tree model
    Numeric variables: 3
    Categorical variables: 5
    Ordinal variables: 3

Consists of 220 clusters, spread across 16 tree branches

Examining the results more closely

## Double-check the data (last 2 outliers)
titanic[c(1147, 1164), ]

##    Pclass Survived    Sex Age SibSp Parch   Fare Cabin Embarked Boat Body
## 1:      3       No Female  39     0     5 29.125  <NA>        Q <NA>  327
## 2:      3       No   Male  NA     0     0 24.150  <NA>        Q <NA>   NA

## Distribution of the group from which those two outliers were flagged
titanic[
    Pclass == 3 &
    SibSp == 0 &
    Embarked == "Q"
][
    , Fare
] |>
    hist(breaks = 100, col = "navy", xlab="Fare",
         main="Distribution of Fare within cluster")

## Get the outliers in a manipulable format
predict(otree, titanic, outliers_print = 0)[[1147]]

$suspicous_value
$suspicous_value$column
[1] "Fare"

$suspicous_value$value
[1] 29.125

$suspicous_value$decimals
[1] 0


$group_statistics
$group_statistics$upper_thr
[1] 15.5

$group_statistics$pct_below
[1] 0.9784946

$group_statistics$mean
[1] 7.886953

$group_statistics$sd
[1] 1.173321

$group_statistics$n_obs
[1] 91


$conditions
$conditions[[1]]
$conditions[[1]]$column
[1] "Embarked"

$conditions[[1]]$value_this
[1] "Q"

$conditions[[1]]$comparison
[1] "="

$conditions[[1]]$value_comp
[1] "Q"


$conditions[[2]]
$conditions[[2]]$column
[1] "Pclass"

$conditions[[2]]$value_this
[1] "3"

$conditions[[2]]$comparison
[1] "in"

$conditions[[2]]$value_comp
[1] "3"


$conditions[[3]]
$conditions[[3]]$column
[1] "SibSp"

$conditions[[3]]$value_this
[1] "0"

$conditions[[3]]$comparison
[1] "in"

$conditions[[3]]$value_comp
[1] "0"


$conditions[[4]]
$conditions[[4]]$column
[1] "Pclass"

$conditions[[4]]$value_this
[1] "3"

$conditions[[4]]$comparison
[1] "in"

$conditions[[4]]$value_comp
[1] "2" "3"



$tree_depth
[1] 4

$uses_NA_branch
[1] FALSE

$outlier_score
[1] 0.003805098

## To programatically get all the outliers that were flagged
pred <- predict(otree, titanic, outliers_print = 0)
only_flagged <- pred[!is.na(sapply(pred, function(x) x$outlier_score))]

## To print selected rows only
print(pred, only_these_rows = 1147)

Reporting top 1 outliers [out of 1 found]

row [1147] - suspicious column: [Fare] - suspicious value: [29.12]
    distribution: 97.849% <= 15.50 - [mean: 7.89] - [sd: 1.17] - [norm. obs: 91]
    given:
        [Pclass] = [3]
        [SibSp] = [0]
        [Embarked] = [Q]

Trying different hyperparameters

## In order to flag more outliers, one can also experiment
## with lowering the threshold hyperparameters
outlier.tree(titanic, z_outlier = 6., outliers_print = 5)

Reporting top 5 outliers [out of 20 found]

row [364] - suspicious column: [Fare] - suspicious value: [0.00]
    distribution: 98.555% >= 3.17 - [mean: 11.66] - [sd: 9.02] - [norm. obs: 682]
    given:
        [Pclass] in [2, 3] (value: 2)
        [SibSp] = [0]


row [385] - suspicious column: [Fare] - suspicious value: [0.00]
    distribution: 98.555% >= 3.17 - [mean: 11.66] - [sd: 9.02] - [norm. obs: 682]
    given:
        [Pclass] in [2, 3] (value: 2)
        [SibSp] = [0]


row [411] - suspicious column: [Fare] - suspicious value: [0.00]
    distribution: 98.555% >= 3.17 - [mean: 11.66] - [sd: 9.02] - [norm. obs: 682]
    given:
        [Pclass] in [2, 3] (value: 2)
        [SibSp] = [0]


row [474] - suspicious column: [Fare] - suspicious value: [0.00]
    distribution: 98.555% >= 3.17 - [mean: 11.66] - [sd: 9.02] - [norm. obs: 682]
    given:
        [Pclass] in [2, 3] (value: 2)
        [SibSp] = [0]


row [529] - suspicious column: [Fare] - suspicious value: [0.00]
    distribution: 98.555% >= 3.17 - [mean: 11.66] - [sd: 9.02] - [norm. obs: 682]
    given:
        [Pclass] in [2, 3] (value: 2)
        [SibSp] = [0]

Outlier Tree model
    Numeric variables: 3
    Categorical variables: 5
    Ordinal variables: 3

Consists of 216 clusters, spread across 16 tree branches

## One can also lower the gain threshold, but this tends
## to result in more spurious outliers which come from
## not-so-good splits (not recommended)
outlier.tree(titanic, z_outlier = 6., min_gain = 1e-6, outliers_print = 5)

Reporting top 5 outliers [out of 27 found]

row [546] - suspicious column: [SibSp] - suspicious value: [3]
    distribution: 99.701% in [0, 1, 2, 5, 8]
    ( [norm. obs: 999] - [prior_prob: 1.528%] - [next smallest: 2.595%] )
    given:
        [Parch] = [0]


row [657] - suspicious column: [SibSp] - suspicious value: [3]
    distribution: 99.701% in [0, 1, 2, 5, 8]
    ( [norm. obs: 999] - [prior_prob: 1.528%] - [next smallest: 2.595%] )
    given:
        [Parch] = [0]


row [1275] - suspicious column: [SibSp] - suspicious value: [3]
    distribution: 99.701% in [0, 1, 2, 5, 8]
    ( [norm. obs: 999] - [prior_prob: 1.528%] - [next smallest: 2.595%] )
    given:
        [Parch] = [0]


row [364] - suspicious column: [Fare] - suspicious value: [0.00]
    distribution: 98.555% >= 3.17 - [mean: 11.66] - [sd: 9.02] - [norm. obs: 682]
    given:
        [Pclass] in [2, 3] (value: 2)
        [SibSp] = [0]


row [385] - suspicious column: [Fare] - suspicious value: [0.00]
    distribution: 98.555% >= 3.17 - [mean: 11.66] - [sd: 9.02] - [norm. obs: 682]
    given:
        [Pclass] in [2, 3] (value: 2)
        [SibSp] = [0]

Outlier Tree model
    Numeric variables: 3
    Categorical variables: 5
    Ordinal variables: 3

Consists of 285 clusters, spread across 23 tree branches