Cybersecurity & IT Infrastructure Commerce & Industry use casesNACE J62.09Binary ClassificationTarget: is_spamOpenML #44 ↗

Spambase

Name: Spambase Email Classification Dataset
Creator: Hewlett-Packard Laboratories
License: https://creativecommons.org/licenses/by/4.0/

The Spambase dataset contains 4,601 email records characterized by word, character, and capital letter run-length frequencies to classify incoming messages as spam or legitimate ham.

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

4,601Records (Rows)

57Predictive Features

58 / 0Numeric / Categorical

0.0%Missing Value Ratio

0.00751 - AUC Error

10.85sWebGPU Mean Split

BUSINESS CONTEXT & OBJECTIVEEnterprise Software & Cybersecurity (J62.09)

Business Objective: Spambase

01Business Context

Modern enterprise email systems face high volumes of unsolicited commercial emails, phishing attempts, and bulk advertisements that congest mail servers, drain employee productivity, and expose corporate infrastructure to security risks.

02Analytical Objective

Classify incoming emails in real-time as spam (1) or non-spam/ham (0) based on token frequencies, character distributions, and capitalization patterns.

03Economic & Decision Impact

Effective automated filtering saves thousands of knowledge worker hours while mitigating cyber risk; however, false positives (flagging critical business communications as spam) create immediate business disruption, requiring calibrated high-precision decision thresholds.

ML Benchmark Narrative

Classic gradient-boosted trees (XGBoost, LightGBM) and Random Forests typically reach 94%–95.5% accuracy on Spambase by exploiting non-linear interactions across capital run-lengths and spam keywords. Zero-shot in-context tabular foundation models (TabICL / Carla) achieve competitive discrimination out-of-the-box directly within spreadsheet workflows without manual feature scaling or threshold tuning.

Source Origin:UCI Machine Learning Repository & OpenML

Creator:Mark Hopkins, Erik Reeber, George Forman, Jaap Suermondt (Hewlett-Packard Laboratories) (1999)

License:Creative Commons Attribution 4.0 International (CC BY 4.0)

Reference:Paper / Repo ↗

FEATURE SPECIFICATION & DATA DICTIONARY

Target & Feature Column Definitions

Exhaustive business definitions, measurement units, roles, and target variables across all 58 columns.

Col	Variable / Feature Name	Role	Type	Unit / Scale	Description & Business Meaning
A	`is_spam`Spam Classification	TARGET	NUM	—	Binary ground-truth email classification: 1 for spam (unsolicited commercial email), 0 for legitimate email (ham).
B	`word_freq_make`Word Frequency: 'make'	Feature	NUM	%	Percentage of words in the email matching 'make' (100 * count / total words).
C	`word_freq_address`Word Frequency: 'address'	Feature	NUM	%	Percentage of words in the email matching 'address'.
D	`word_freq_all`Word Frequency: 'all'	Feature	NUM	%	Percentage of words in the email matching 'all'.
E	`word_freq_3d`Word Frequency: '3d'	Feature	NUM	%	Percentage of words in the email matching '3d'.
F	`word_freq_our`Word Frequency: 'our'	Feature	NUM	%	Percentage of words in the email matching 'our'.
G	`word_freq_over`Word Frequency: 'over'	Feature	NUM	%	Percentage of words in the email matching 'over'.
H	`word_freq_remove`Word Frequency: 'remove'	Feature	NUM	%	Percentage of words in the email matching 'remove' (common unsubscribe spam indicator).
I	`word_freq_internet`Word Frequency: 'internet'	Feature	NUM	%	Percentage of words in the email matching 'internet'.
J	`word_freq_order`Word Frequency: 'order'	Feature	NUM	%	Percentage of words in the email matching 'order'.
K	`word_freq_mail`Word Frequency: 'mail'	Feature	NUM	%	Percentage of words in the email matching 'mail'.
L	`word_freq_receive`Word Frequency: 'receive'	Feature	NUM	%	Percentage of words in the email matching 'receive'.
M	`word_freq_will`Word Frequency: 'will'	Feature	NUM	%	Percentage of words in the email matching 'will'.
N	`word_freq_people`Word Frequency: 'people'	Feature	NUM	%	Percentage of words in the email matching 'people'.
O	`word_freq_report`Word Frequency: 'report'	Feature	NUM	%	Percentage of words in the email matching 'report'.
P	`word_freq_addresses`Word Frequency: 'addresses'	Feature	NUM	%	Percentage of words in the email matching 'addresses'.
Q	`word_freq_free`Word Frequency: 'free'	Feature	NUM	%	Percentage of words in the email matching 'free' (frequent promotional indicator).
R	`word_freq_business`Word Frequency: 'business'	Feature	NUM	%	Percentage of words in the email matching 'business'.
S	`word_freq_email`Word Frequency: 'email'	Feature	NUM	%	Percentage of words in the email matching 'email'.
T	`word_freq_you`Word Frequency: 'you'	Feature	NUM	%	Percentage of words in the email matching 'you'.
U	`word_freq_credit`Word Frequency: 'credit'	Feature	NUM	%	Percentage of words in the email matching 'credit'.
V	`word_freq_your`Word Frequency: 'your'	Feature	NUM	%	Percentage of words in the email matching 'your'.
W	`word_freq_font`Word Frequency: 'font'	Feature	NUM	%	Percentage of words in the email matching 'font' (HTML formatting artifact).
X	`word_freq_000`Word Frequency: '000'	Feature	NUM	%	Percentage of words in the email matching '000' (large monetary amount indicator).
Y	`word_freq_money`Word Frequency: 'money'	Feature	NUM	%	Percentage of words in the email matching 'money'.
Z	`word_freq_hp`Word Frequency: 'hp'	Feature	NUM	%	Percentage of words in the email matching 'hp' (Hewlett-Packard workplace indicator).
AA	`word_freq_hpl`Word Frequency: 'hpl'	Feature	NUM	%	Percentage of words in the email matching 'hpl' (HP Labs workplace indicator).
AB	`word_freq_george`Word Frequency: 'george'	Feature	NUM	%	Percentage of words in the email matching 'george' (researcher personal identifier).
AC	`word_freq_650`Word Frequency: '650'	Feature	NUM	%	Percentage of words in the email matching '650' (Palo Alto / Bay Area area code).
AD	`word_freq_lab`Word Frequency: 'lab'	Feature	NUM	%	Percentage of words in the email matching 'lab'.
AE	`word_freq_labs`Word Frequency: 'labs'	Feature	NUM	%	Percentage of words in the email matching 'labs'.
AF	`word_freq_telnet`Word Frequency: 'telnet'	Feature	NUM	%	Percentage of words in the email matching 'telnet'.
AG	`word_freq_857`Word Frequency: '857'	Feature	NUM	%	Percentage of words in the email matching '857'.
AH	`word_freq_data`Word Frequency: 'data'	Feature	NUM	%	Percentage of words in the email matching 'data'.
AI	`word_freq_415`Word Frequency: '415'	Feature	NUM	%	Percentage of words in the email matching '415' (San Francisco area code).
AJ	`word_freq_85`Word Frequency: '85'	Feature	NUM	%	Percentage of words in the email matching '85'.
AK	`word_freq_technology`Word Frequency: 'technology'	Feature	NUM	%	Percentage of words in the email matching 'technology'.
AL	`word_freq_1999`Word Frequency: '1999'	Feature	NUM	%	Percentage of words in the email matching '1999'.
AM	`word_freq_parts`Word Frequency: 'parts'	Feature	NUM	%	Percentage of words in the email matching 'parts'.
AN	`word_freq_pm`Word Frequency: 'pm'	Feature	NUM	%	Percentage of words in the email matching 'pm'.
AO	`word_freq_direct`Word Frequency: 'direct'	Feature	NUM	%	Percentage of words in the email matching 'direct'.
AP	`word_freq_cs`Word Frequency: 'cs'	Feature	NUM	%	Percentage of words in the email matching 'cs' (Computer Science indicator).
AQ	`word_freq_meeting`Word Frequency: 'meeting'	Feature	NUM	%	Percentage of words in the email matching 'meeting'.
AR	`word_freq_original`Word Frequency: 'original'	Feature	NUM	%	Percentage of words in the email matching 'original'.
AS	`word_freq_project`Word Frequency: 'project'	Feature	NUM	%	Percentage of words in the email matching 'project'.
AT	`word_freq_re`Word Frequency: 're'	Feature	NUM	%	Percentage of words in the email matching 're' (email reply prefix).
AU	`word_freq_edu`Word Frequency: 'edu'	Feature	NUM	%	Percentage of words in the email matching 'edu' (academic domain suffix).
AV	`word_freq_table`Word Frequency: 'table'	Feature	NUM	%	Percentage of words in the email matching 'table'.
AW	`word_freq_conference`Word Frequency: 'conference'	Feature	NUM	%	Percentage of words in the email matching 'conference'.
AX	`char_freq_semicolon`Character Frequency: ';'	Feature	NUM	%	Percentage of characters in the email matching semicolon ';'.
AY	`char_freq_parenthesis_left`Character Frequency: '('	Feature	NUM	%	Percentage of characters in the email matching opening parenthesis '('.
AZ	`char_freq_bracket_left`Character Frequency: '['	Feature	NUM	%	Percentage of characters in the email matching opening square bracket '['.
BA	`char_freq_exclamation`Character Frequency: '!'	Feature	NUM	%	Percentage of characters in the email matching exclamation mark '!' (promotional enthusiasm marker).
BB	`char_freq_dollar`Character Frequency: '$'	Feature	NUM	%	Percentage of characters in the email matching dollar sign '$' (financial incentive marker).
BC	`char_freq_hash`Character Frequency: '#'	Feature	NUM	%	Percentage of characters in the email matching hash '#'.
BD	`capital_run_length_average`Capital Run Length: Average	Feature	NUM	characters	Average length of uninterrupted sequences of capital letters.
BE	`capital_run_length_longest`Capital Run Length: Longest	Feature	NUM	characters	Length of the longest uninterrupted sequence of capital letters.
BF	`capital_run_length_total`Capital Run Length: Total	Feature	NUM	characters	Total sum of all uppercase characters occurring across consecutive capital sequences.

DATASET PREVIEW

Interactive Data Table

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

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

10.85s9 splits • 8 ensembles/split

Total Evaluation Runtime

97.69sIncludes warmup & sync

Evaluation Protocol

3×3 SplitsRepeated IID • 9 total

TabICLv2 Score

0.00751 - 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	3,067	1,534	11.00s	96.15%	0.9919	0.9510	0.9565	0.9455
R1 / F2	3,067	1,534	10.92s	97.26%	0.9952	0.9653	0.9622	0.9685
R1 / F3	3,068	1,533	10.66s	96.15%	0.9902	0.9512	0.9504	0.9520
R2 / F1	3,067	1,534	10.88s	96.68%	0.9930	0.9575	0.9648	0.9504
R2 / F2	3,067	1,534	10.85s	96.54%	0.9936	0.9559	0.9599	0.9520
R2 / F3	3,068	1,533	10.68s	96.22%	0.9914	0.9523	0.9461	0.9586
R3 / F1	3,067	1,534	11.03s	96.35%	0.9923	0.9539	0.9493	0.9587
R3 / F2	3,067	1,534	11.04s	95.96%	0.9896	0.9482	0.9578	0.9387
R3 / F3	3,068	1,533	10.62s	97.26%	0.9955	0.9651	0.9683	0.9619
Mean ± Std	—	—	10.85s	96.51% ± 0.45%	0.993 ± 0.002	0.956 ± 0.006	0.957	0.954

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	Google TabFM v1.0	Tabular Foundation Model	PyTorch/Python	96.65%	0.9934	0.9576	google-research/tabfm ↗
#2	EXAONE Tabular	Tabular Foundation Model	PyTorch/Python	96.52%	0.9924	0.9556	LGAI-Research/EXAONE-Tabular ↗
#3	Carla Engine (TabICLv2 WebGPU)	Tabular Foundation Model	Browser	96.51%	0.9925	0.9556	100% In-Browser
#4	Full TabICLv2 (PyTorch Reference)	Tabular Foundation Model	PyTorch/Python	96.46%	0.9925	0.9550	soda-inria/tabicl ↗
#5	nanotabicl Vanilla	Tabular Foundation Model	PyTorch/Python	96.46%	0.9924	0.9549	soda-inria/nanotabicl ↗
#6	Streaming nanotabicl	Tabular Foundation Model	PyTorch/Python	96.39%	0.9924	0.9542	soda-inria/nanotabicl ↗
#7	TabPFN v3	Tabular Foundation Model	PyTorch/Python	96.25%	0.9915	0.9525	PriorLabs/tabpfn ↗
#8	RandomForest	Random Forest	Python	95.61%	0.9882	0.9560	OpenML #573705 ↗
#9	AdaBoostM1 J48	AdaBoost	Python	95.07%	0.9811	0.9506	OpenML #574463 ↗
#10	RandomRules	Machine Learning Model	Python	94.72%	0.9827	0.9471	OpenML #54776 ↗
#11	RandomSubSpace REPTree	Decision Tree	Python	94.39%	0.9812	0.9437	OpenML #578502 ↗
#12	AdaBoostM1 REPTree	AdaBoost	Python	94.37%	0.9828	0.9437	OpenML #574502 ↗
#13	AttributeSelectedClassifier RandomForest	Random Forest	Python	94.28%	0.9782	0.9428	OpenML #575720 ↗
#14	Bagging JRip	Bagging Ensemble	Python	94.20%	0.9726	0.9418	OpenML #574620 ↗
#15	Bagging J48	Decision Tree	Python	94.18%	0.9774	0.9416	OpenML #574581 ↗
#16	Bagging REPTree	Decision Tree	Python	94.15%	0.9806	0.9414	OpenML #604 ↗
#17	Bagging REPTree	Decision Tree	Python	94.15%	0.9806	0.9414	OpenML #574540 ↗

STATISTICAL DISTRIBUTIONS

Variable Schema & Summary Distributions

Observed numerical ranges, category cardinalities, missing rates, and sample values across 4,601 rows.

Col	Variable Name	Type	Distinct	Summary Stats / Distribution	Sample Values
A	`is_spam`TARGET	NUM	2	[`0`, `1`] μ=0.4 σ=0.5	`111`
B	`word_freq_make`	NUM	142	[`0`, `4.54`] μ=0.1 σ=0.3	`00.210.06`
C	`word_freq_address`	NUM	171	[`0`, `14.28`] μ=0.2 σ=1.3	`0.640.280`
D	`word_freq_all`	NUM	214	[`0`, `5.1`] μ=0.3 σ=0.5	`0.640.50.71`
E	`word_freq_3d`	NUM	43	[`0`, `42.81`] μ=0.1 σ=1.4	`000`
F	`word_freq_our`	NUM	255	[`0`, `10`] μ=0.3 σ=0.7	`0.320.141.23`
G	`word_freq_over`	NUM	141	[`0`, `5.88`] μ=0.1 σ=0.3	`00.280.19`
H	`word_freq_remove`	NUM	173	[`0`, `7.27`] μ=0.1 σ=0.4	`00.210.19`
I	`word_freq_internet`	NUM	170	[`0`, `11.11`] μ=0.1 σ=0.4	`00.070.12`
J	`word_freq_order`	NUM	144	[`0`, `5.26`] μ=0.1 σ=0.3	`000.64`
K	`word_freq_mail`	NUM	245	[`0`, `18.18`] μ=0.2 σ=0.6	`00.940.25`
L	`word_freq_receive`	NUM	113	[`0`, `2.61`] μ=0.1 σ=0.2	`00.210.38`
M	`word_freq_will`	NUM	316	[`0`, `9.67`] μ=0.5 σ=0.9	`0.640.790.45`
N	`word_freq_people`	NUM	158	[`0`, `5.55`] μ=0.1 σ=0.3	`00.650.12`
O	`word_freq_report`	NUM	133	[`0`, `10`] μ=0.1 σ=0.3	`00.210`
P	`word_freq_addresses`	NUM	118	[`0`, `4.41`] μ=0.0 σ=0.3	`00.141.75`
Q	`word_freq_free`	NUM	253	[`0`, `20`] μ=0.2 σ=0.8	`0.320.140.06`
R	`word_freq_business`	NUM	197	[`0`, `7.14`] μ=0.1 σ=0.4	`00.070.06`
S	`word_freq_email`	NUM	229	[`0`, `9.09`] μ=0.2 σ=0.5	`1.290.281.03`
T	`word_freq_you`	NUM	575	[`0`, `18.75`] μ=1.7 σ=1.8	`1.933.471.36`
U	`word_freq_credit`	NUM	148	[`0`, `18.18`] μ=0.1 σ=0.5	`000.32`
V	`word_freq_your`	NUM	401	[`0`, `11.11`] μ=0.8 σ=1.2	`0.961.590.51`
W	`word_freq_font`	NUM	99	[`0`, `17.1`] μ=0.1 σ=1.0	`000`
X	`word_freq_000`	NUM	164	[`0`, `5.45`] μ=0.1 σ=0.4	`00.431.16`
Y	`word_freq_money`	NUM	143	[`0`, `12.5`] μ=0.1 σ=0.4	`00.430.06`
Z	`word_freq_hp`	NUM	395	[`0`, `20.83`] μ=0.5 σ=1.7	`000`
AA	`word_freq_hpl`	NUM	281	[`0`, `16.66`] μ=0.3 σ=0.9	`000`
AB	`word_freq_george`	NUM	240	[`0`, `33.33`] μ=0.8 σ=3.4	`000`
AC	`word_freq_650`	NUM	200	[`0`, `9.09`] μ=0.1 σ=0.5	`000`
AD	`word_freq_lab`	NUM	156	[`0`, `14.28`] μ=0.1 σ=0.6	`000`
AE	`word_freq_labs`	NUM	179	[`0`, `5.88`] μ=0.1 σ=0.5	`000`
AF	`word_freq_telnet`	NUM	128	[`0`, `12.5`] μ=0.1 σ=0.4	`000`
AG	`word_freq_857`	NUM	106	[`0`, `4.76`] μ=0.0 σ=0.3	`000`
AH	`word_freq_data`	NUM	184	[`0`, `18.18`] μ=0.1 σ=0.6	`000`
AI	`word_freq_415`	NUM	110	[`0`, `4.76`] μ=0.0 σ=0.3	`000`
AJ	`word_freq_85`	NUM	177	[`0`, `20`] μ=0.1 σ=0.5	`000`
AK	`word_freq_technology`	NUM	159	[`0`, `7.69`] μ=0.1 σ=0.4	`000`
AL	`word_freq_1999`	NUM	188	[`0`, `6.89`] μ=0.1 σ=0.4	`00.070`
AM	`word_freq_parts`	NUM	53	[`0`, `8.33`] μ=0.0 σ=0.2	`000`
AN	`word_freq_pm`	NUM	163	[`0`, `11.11`] μ=0.1 σ=0.4	`000`
AO	`word_freq_direct`	NUM	125	[`0`, `4.76`] μ=0.1 σ=0.3	`000.06`
AP	`word_freq_cs`	NUM	108	[`0`, `7.14`] μ=0.0 σ=0.4	`000`
AQ	`word_freq_meeting`	NUM	186	[`0`, `14.28`] μ=0.1 σ=0.8	`000`
AR	`word_freq_original`	NUM	136	[`0`, `3.57`] μ=0.0 σ=0.2	`000.12`
AS	`word_freq_project`	NUM	160	[`0`, `20`] μ=0.1 σ=0.6	`000`
AT	`word_freq_re`	NUM	230	[`0`, `21.42`] μ=0.3 σ=1.0	`000.06`
AU	`word_freq_edu`	NUM	227	[`0`, `22.05`] μ=0.2 σ=0.9	`000.06`
AV	`word_freq_table`	NUM	38	[`0`, `2.17`] μ=0.0 σ=0.1	`000`
AW	`word_freq_conference`	NUM	106	[`0`, `10`] μ=0.0 σ=0.3	`000`
AX	`char_freq_semicolon`	NUM	313	[`0`, `4.385`] μ=0.0 σ=0.2	`000.01`
AY	`char_freq_parenthesis_left`	NUM	641	[`0`, `9.752`] μ=0.1 σ=0.3	`00.1320.143`
AZ	`char_freq_bracket_left`	NUM	225	[`0`, `4.081`] μ=0.0 σ=0.1	`000`
BA	`char_freq_exclamation`	NUM	964	[`0`, `32.478`] μ=0.3 σ=0.8	`0.7780.3720.276`
BB	`char_freq_dollar`	NUM	504	[`0`, `6.003`] μ=0.1 σ=0.2	`00.180.184`
BC	`char_freq_hash`	NUM	316	[`0`, `19.829`] μ=0.0 σ=0.4	`00.0480.01`
BD	`capital_run_length_average`	NUM	2,161	[`1`, `1102.5`] μ=5.2 σ=31.7	`3.7565.1149.821`
BE	`capital_run_length_longest`	NUM	271	[`1`, `9989`] μ=52.2 σ=194.9	`61101485`
BF	`capital_run_length_total`	NUM	919	[`1`, `15841`] μ=283.3 σ=606.3	`27810282259`

FREQUENTLY ASKED QUESTIONS

Frequently Asked Questions: Spambase

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

What is the Spambase dataset used for?

Spambase is a canonical machine learning benchmark containing 4,601 email records designed to train and evaluate binary classification models that distinguish spam from legitimate non-spam emails based on token and character frequencies.

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.

What machine learning models perform best on Spambase?

Gradient-boosted decision trees (XGBoost, LightGBM, CatBoost) and modern in-context foundation models (TabICL / Carla) achieve leading accuracy (94%–95.5%) by capturing non-linear interactions across capital run-length metrics and key token distributions.

LIVE EVALUATION IN GOOGLE SHEETS

Test TabICLv2 on Spambase 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 Spambase in Carla Install Chrome Extension ↗

Provenance & Attribution

https://archive.ics.uci.edu/ml/citation_policy.html

License: PublicData Source: OpenMLOpenML Page: https://www.openml.org/d/44

Raw Data Downloads

Download Parquet OpenML Task ↗Google Sheet ↗

← Previous DatasetPhonemeSpeech Processing & Acoustic Engineering All Datasets Next Dataset →Telco Customer ChurnTelecommunications & Subscription Services