Implementation:Online ml River Datasets Keystroke

Knowledge Sources	Online_ml_River
Domains	Online_Learning, Datasets, Multi_Class_Classification, Biometrics
Last Updated	2026-02-08 16:00 GMT

Overview

Concrete dataset for multi-class classification provided by the River library.

Description

CMU keystroke dataset. Users are tasked to type in a password. The task is to determine which user is typing in the password based on their keystroke dynamics.

The only difference with the original dataset is that the "sessionIndex" and "rep" attributes have been dropped.

This dataset contains 20,400 samples with 31 features across 51 classes for multi-class classification tasks.

Usage

This dataset is useful for:

• Biometric authentication and user identification
• Behavioral biometrics research
• Multi-class classification with moderate number of classes
• Keystroke dynamics analysis

Code Reference

Source Location

• Repository: Online_ml_River
• File: river/datasets/keystroke.py

Signature

class Keystroke(base.RemoteDataset):
    def __init__(self):
        super().__init__(
            n_samples=20_400,
            n_features=31,
            n_classes=51,
            task=base.MULTI_CLF,
            url="http://www.cs.cmu.edu/~keystroke/DSL-StrongPasswordData.csv",
            size=4_669_935,
            filename="DSL-StrongPasswordData.csv",
            unpack=False,
        )

    def _iter(self):
        converters = {
            "H.period": float,
            "DD.period.t": float,
            "UD.period.t": float,
            # ... (31 keystroke timing features)
        }
        return stream.iter_csv(
            self.path,
            target="subject",
            converters=converters,
            drop=["sessionIndex", "rep"],
        )

Import

from river import datasets
dataset = datasets.Keystroke()

I/O Contract

Inputs

Name	Type	Required	Description
(none)	—	—	No parameters needed

Outputs

Name	Type	Description
iter()	tuple(dict, str)	Yields (features_dict, target) pairs where target is the subject/user ID

Dataset Properties

Property	Value
Number of samples	20,400
Number of features	31
Number of classes	51
Task	Multi-class classification
Format	CSV
Size	4,669,935 bytes

Features

The dataset includes 31 keystroke timing features representing:

• Hold time (H): Duration a key is pressed
• Down-Down time (DD): Time between pressing two consecutive keys
• Up-Down time (UD): Time between releasing one key and pressing the next

Features capture typing dynamics for the password ".tie5Roanl" including:

• Individual key hold times (H.period, H.t, H.i, etc.)
• Transition times between keys (DD.period.t, UD.period.t, etc.)

Usage Examples

from river import datasets

dataset = datasets.Keystroke()
for x, y in dataset:
    print(x, y)
    break

References

• Keystroke Dynamics - Benchmark Data Set

Related Pages

• Environment:Online_ml_River_Python_Runtime_Environment