Maritime Safety & Actuarial Risk Analysis Public Services & Risk use casesNACE H50.10Binary ClassificationTarget: survivedOpenML #40945 ↗

Titanic

Canonical 1,309-record benchmark dataset describing Titanic passenger demographics, socio-economic markers, and cabin locations to predict binary survival outcomes.

✨ Try in Carla Install Chrome Extension ↗100% In-Browser WebGPU • Zero Cloud Upload

1,309Records (Rows)

11Predictive Features

7 / 7Numeric / Categorical

21.0%Missing Value Ratio

0.12291 - AUC Error

1.35sWebGPU Mean Split

BUSINESS CONTEXT & OBJECTIVEMaritime Transportation & Insurance Underwriting (H50.10)

Business Objective: Titanic

01Business Context

Passenger evacuation protocols and maritime safety logistics during mass-casualty emergency operations, historically governed by triage heuristics such as 'women and children first' and stratified socio-economic access to survival resources.

02Analytical Objective

Predict the binary survival probability of individual passengers based on socio-economic tier, demographic profile, cabin positioning, and family accompaniment.

03Economic & Decision Impact

Enables historical casualty auditing, safety triage simulation, and actuarial disaster loss modeling where false positives (misclassifying deceased passengers as survivors) distort survival resource allocations and emergency egress planning.

ML Benchmark Narrative

Widely utilized as the definitive baseline for tabular binary classification, traditional gradient-boosted decision trees (XGBoost, LightGBM) and Random Forests routinely achieve 0.80–0.84 ROC-AUC on clean non-leaking feature subsets (pclass, sex, age, sibsp, parch, fare, embarked). Foundation models like TabICL and Carla demonstrate robust zero-shot in-context inference directly on raw spreadsheet rows without explicit imputation pipelines or one-hot categorical encoding.

Source Origin:Vanderbilt University Department of Biostatistics / Encyclopedia Titanica

Creator:Frank E. Harrell Jr. and Thomas Cason (1994)

License:Open Data Commons Public Domain Dedication (CC0 / Public Domain)

Reference:Paper / Repo ↗

FEATURE SPECIFICATION & DATA DICTIONARY

Target & Feature Column Definitions

Exhaustive business definitions, measurement units, roles, and target variables across all 14 columns.

Col	Variable / Feature Name	Role	Type	Unit / Scale	Description & Business Meaning
A	`survived`Survival Status	TARGET	NUM	—	Binary passenger outcome indicating whether the individual survived the disaster (0 = Did not survive / Deceased, 1 = Survived).
B	`pclass`Passenger Class	Feature	NUM	—	Ticket class representing socio-economic status (1 = 1st Class / Upper Tier, 2 = 2nd Class / Middle Tier, 3 = 3rd Class / Lower Tier).
C	`name`Passenger Name	Feature	CAT	—	Full formal name and honorific title (e.g. Mr., Mrs., Miss, Master, Dr., Rev.) of the passenger.
D	`sex`Sex	Feature	CAT	—	Biological sex / gender of the passenger (female, male).
E	`age`Age	Feature	NUM	years	Age of the passenger in years (fractional for infants under 1 year old; missing values represent unknown ages).
F	`sibsp`Siblings / Spouses Aboard	Feature	NUM	count	Total count of siblings (brother, sister, stepbrother, stepsister) or spouses (husband, wife) traveling with the passenger.
G	`parch`Parents / Children Aboard	Feature	NUM	count	Total count of parents (mother, father) or children (daughter, son, stepdaughter, stepson) traveling with the passenger.
H	`ticket`Ticket Number	Feature	CAT	—	Alphanumeric ticket booking identifier or serial reference assigned to the passenger.
I	`fare`Passenger Fare	Feature	NUM	GBP	Total passenger fare paid in pre-1970 British Pounds (£) for boarding the voyage.
J	`cabin`Cabin Number	Feature	CAT	—	Assigned deck and stateroom cabin identifier (e.g. B5, C22 C26). Deck letter reflects vertical proximity to boat deck.
K	`embarked`Port of Embarkation	Feature	CAT	—	Port where the passenger boarded the vessel (C = Cherbourg, Q = Queenstown, S = Southampton).
L	`boat`Lifeboat Identifier	LEAKAGE	CAT	—	Identification number or letter of the rescue lifeboat used by surviving passengers (post-event indicator). TARGET LEAKAGE RISK(post event) The lifeboat identifier indicates which rescue boat a passenger boarded during or after the disaster. This feature is recorded post-event and provides direct leakage of survival status, as having an assigned lifeboat strongly determines that the passenger survived. • Excluded during model training & zero-shot inference to avoid artificially inflated metric scores.
L	`homedest`Home / Destination	Feature	CAT	—	Geographic origin / home city and ultimate travel destination intended by the passenger.
M	`body`Body Identification Number	LEAKAGE	NUM	—	Coroner recovery identification index assigned to deceased passenger bodies retrieved from the ocean. TARGET LEAKAGE RISK(post event) The body identification number is assigned by recovery crews and coroners to deceased passengers whose bodies were retrieved from the ocean after the sinking. This is a post-event feature that deterministically indicates the passenger did not survive. • Excluded during model training & zero-shot inference to avoid artificially inflated metric scores.

DATASET PREVIEW

Interactive Data Table

Explore rows, feature values, and target labels for Titanic.

Download ParquetHF Viewer ↗

LOCAL EXECUTION TELEMETRY

TabICLv2 WebGPU Benchmark Results

Complete performance metrics on persisted IID or non-IID evaluation splits, executed 100% locally in the browser sandbox.

Mean Split Latency

1.35s30 splits • 8 ensembles/split

Total Evaluation Runtime

40.50sIncludes warmup & sync

Evaluation Protocol

10×3 SplitsRepeated IID • 30 total

TabICLv2 Score

0.12291 - AUC Error

Evaluation Split	Train Rows	Test Rows	Split Duration?Wall-clock inference time taken for this evaluation split running 100% locally via WebGPU.Learn more →	Accuracy?Proportion of correct test predictions across all classes.Google ML Guide & Details →	ROC-AUC?Area under the ROC curve evaluating ranking capability across all thresholds.Google ML Guide & Details →	F1 Score?Harmonic mean of Precision and Recall, robust against class imbalance.Learn more →	Precision?Proportion of predicted positives that were actually positive.Google ML Guide & Details →	Recall?Proportion of actual positive cases successfully captured.Google ML Guide & Details →
R1 / F1	872	437	1.42s	79.86%	0.8519	0.7233	0.7616	0.6886
R1 / F2	873	436	1.41s	81.65%	0.8707	0.7403	0.8085	0.6826
R1 / F3	873	436	1.34s	82.57%	0.8780	0.7610	0.7961	0.7289
R2 / F1	872	437	1.35s	79.18%	0.8708	0.7036	0.7714	0.6467
R2 / F2	873	436	1.34s	81.19%	0.8488	0.7303	0.8102	0.6647
R2 / F3	873	436	1.33s	85.09%	0.8950	0.7855	0.8686	0.7169
R3 / F1	872	437	1.34s	85.13%	0.8981	0.7766	0.9113	0.6766
R3 / F2	873	436	1.34s	81.65%	0.8630	0.7403	0.8085	0.6826
R3 / F3	873	436	1.35s	82.57%	0.8775	0.7669	0.7813	0.7530
R4 / F1	872	437	1.35s	83.07%	0.8560	0.7566	0.8394	0.6886
R4 / F2	873	436	1.35s	82.34%	0.8750	0.7556	0.8041	0.7126
R4 / F3	873	436	1.34s	81.42%	0.8827	0.7344	0.8058	0.6747
R5 / F1	872	437	1.34s	79.86%	0.8564	0.6986	0.8160	0.6108
R5 / F2	873	436	1.37s	85.32%	0.8999	0.7987	0.8411	0.7605
R5 / F3	873	436	1.34s	82.57%	0.8903	0.7580	0.8041	0.7169
R6 / F1	872	437	1.35s	81.01%	0.8645	0.7492	0.7561	0.7425
R6 / F2	873	436	1.35s	83.26%	0.8948	0.7638	0.8310	0.7066
R6 / F3	873	436	1.38s	80.96%	0.8534	0.7167	0.8268	0.6325
R7 / F1	872	437	1.35s	83.07%	0.8867	0.7613	0.8252	0.7066
R7 / F2	873	436	1.34s	81.42%	0.8655	0.7344	0.8116	0.6707
R7 / F3	873	436	1.34s	81.19%	0.8794	0.7172	0.8387	0.6265
R8 / F1	872	437	1.35s	78.03%	0.8302	0.6842	0.7591	0.6228
R8 / F2	873	436	1.34s	82.34%	0.8814	0.7475	0.8261	0.6826
R8 / F3	873	436	1.33s	85.09%	0.9021	0.7962	0.8301	0.7651
R9 / F1	872	437	1.32s	81.92%	0.8710	0.7358	0.8333	0.6587
R9 / F2	873	436	1.34s	79.59%	0.8459	0.7192	0.7600	0.6826
R9 / F3	873	436	1.36s	87.16%	0.9155	0.8170	0.8929	0.7530
R10 / F1	872	437	1.33s	81.92%	0.8781	0.7443	0.8099	0.6886
R10 / F2	873	436	1.35s	83.49%	0.8989	0.7647	0.8417	0.7006
R10 / F3	873	436	1.34s	83.26%	0.8670	0.7509	0.8661	0.6627
Mean ± Std	—	—	1.35s	82.24% ± 1.97%	0.875 ± 0.019	0.748 ± 0.030	0.818	0.690

UNIFIED BENCHMARK LEADERBOARD

Foundation Models vs Traditional ML Baselines

Side-by-side evaluation of our in-browser Carla engine (WebGPU), open-source tabular foundation models, and OpenML baselines.

Rank	Algorithm / Model	Model Family	Runtime	Predictive Accuracy?Proportion of correct test predictions across cross-validation splits.Google ML Guide & Details →	ROC-AUC?Area under the ROC curve evaluating ranking capability.Learn more →	F-Measure?Harmonic mean of precision and recall.Learn more →	Reference / Repo
#1	EXAONE Tabular	Tabular Foundation Model	PyTorch/Python	82.96%	0.8854	0.7569	LGAI-Research/EXAONE-Tabular ↗
#2	Google TabFM v1.0	Tabular Foundation Model	PyTorch/Python	82.47%	0.8834	0.7509	google-research/tabfm ↗
#3	Streaming nanotabicl	Tabular Foundation Model	PyTorch/Python	82.38%	0.8771	0.7498	soda-inria/nanotabicl ↗
#4	Full TabICLv2 (PyTorch Reference)	Tabular Foundation Model	PyTorch/Python	82.38%	0.8771	0.7498	soda-inria/tabicl ↗
#5	nanotabicl Vanilla	Tabular Foundation Model	PyTorch/Python	82.35%	0.8771	0.7494	soda-inria/nanotabicl ↗
#6	Carla Engine (TabICLv2 WebGPU)	Tabular Foundation Model	Browser	82.24%	0.8749	0.7477	100% In-Browser
#7	TabPFN v3	Tabular Foundation Model	PyTorch/Python	81.13%	0.8640	0.7327	PriorLabs/tabpfn ↗

STATISTICAL DISTRIBUTIONS

Variable Schema & Summary Distributions

Observed numerical ranges, category cardinalities, missing rates, and sample values across 1,309 rows.

Col	Variable Name	Type	Missing	Distinct	Summary Stats / Distribution	Sample Values
A	`survived`TARGET	NUM	0	2	[`0`, `1`] μ=0.4 σ=0.5	`110`
B	`pclass`	NUM	0	3	[`1`, `3`] μ=2.3 σ=0.8	`111`
C	`name`	CAT	0	1,307	—	`Allen, Miss. Elisabeth WaltonAllison, Master. Hudson TrevorAllison, Miss. Helen Loraine`
D	`sex`	CAT	0	2	2 cats: male, female	`femalemalefemale`
E	`age`	NUM	263 (20.09%)	98	[`0.1667`, `80`] μ=29.9 σ=14.4	`290.91672`
F	`sibsp`	NUM	0	7	[`0`, `8`] μ=0.5 σ=1.0	`011`
G	`parch`	NUM	0	8	[`0`, `9`] μ=0.4 σ=0.9	`022`
H	`ticket`	CAT	0	929	—	`24160113781113781`
I	`fare`	NUM	1 (0.08%)	281	[`0`, `512.3292`] μ=33.3 σ=51.7	`211.3375151.55151.55`
J	`cabin`	CAT	1,014 (77.46%)	186	—	`B5C22 C26C22 C26`
K	`embarked`	CAT	2 (0.15%)	3	3 cats: S, C, Q	`SSS`
L	`boat`LEAKAGE	CAT	823 (62.87%)	27	10 cats: 13, C, 15 +7 more	`2113`
L	`homedest`	CAT	564 (43.09%)	369	—	`St Louis, MOMontreal, PQ / Chesterville, ONMontreal, PQ / Chesterville, ON`
M	`body`LEAKAGE	NUM	1,188 (90.76%)	121	[`1`, `328`] μ=160.8 σ=97.3	`13522124`

FREQUENTLY ASKED QUESTIONS

Frequently Asked Questions: Titanic

Common questions regarding the Titanic dataset, machine learning task formulations, and in-browser tabular inference.

What is the Titanic dataset used for?

The Titanic dataset is the premier tabular benchmark for binary classification, used to predict passenger survival based on socio-economic tier, demographic profiles, family accompaniment, and travel attributes.

Can I run zero-server predictions on this dataset in Google Sheets?

Yes, Carla HQ enables zero-server in-context tabular prediction directly in spreadsheets using TabICL foundation models without requiring backend configuration or feature engineering.

What machine learning models perform best on Titanic?

Standard gradient-boosted trees (LightGBM, XGBoost) and Random Forests typically reach 0.80–0.84 ROC-AUC on clean non-leaking feature sets, while tabular foundation models like TabICL deliver high-accuracy zero-shot inference without prior training.

LIVE EVALUATION IN GOOGLE SHEETS

Test TabICLv2 on Titanic Yourself

Open the pre-loaded Google Sheet and let Carla configure the target and task for local, zero-cloud tabular machine learning.

Step 1

Launch Carla

Click once to open the spreadsheet and Carla side panel together.

Step 2

Review the Setup

Carla selects the dataset target and task from this page automatically.

Step 3

Evaluate & Predict

Run predictions and compute metrics with zero server uploads.

✨ Try Titanic in Carla Install Chrome Extension ↗

Provenance & Attribution

OpenML Dataset 40945

License: PublicData Source: OpenMLOpenML Page: https://www.openml.org/d/40945

Raw Data Downloads

Download Parquet OpenML Task ↗Google Sheet ↗

← Previous DatasetTelco Customer ChurnTelecommunications & Subscription Services All Datasets Next Dataset →VehicleAutomotive Engineering & Computer Vision