Sherlock

The dataset is essentially a massive time-series dataset spanning nearly every single kind of software and hardware sensor that can be sampled from a Samsung Galaxy S5 smartphone, without root privileges. The dataset contains over 600 billion data points in over 10 billion data records. Some examples of the sampled sensors are: Resource utilization per running App (CPU, memory, …) Call/SMS logs Location WiFi Signal strength Network statistics And many more… (see the dataset description here) These sensors where sampled as a rate rivaling other similar datasets, some features sampled at a rate of up to once every second! More interestingly, we provide explicit labels (timestamps + descriptions) which capture exactly when malware on the device is performing its malicious activities. With these labels, you can use the dataset as a benchmark for your machine learning algorithms.

本数据集为超大规模时序数据集，覆盖三星Galaxy S5智能手机在无需root权限的前提下可采集的几乎全部软硬件传感器数据。该数据集包含超100亿条数据记录，总计6000亿余个数据点。可采集的传感器示例包括：各正在运行的应用的资源占用情况（CPU、内存等）、通话/短信日志、位置信息、WiFi信号强度、网络统计数据等更多类型（详见本数据集官方描述文档）。该数据集的传感器采样速率可与同类数据集媲美，部分特征的采样频率最高可达每秒一次！更具价值的是，本数据集附带精准标注信息（含时间戳与活动描述），可精确标记设备上恶意软件执行恶意行为的具体时刻。依托这些标注，该数据集可作为机器学习算法的基准测试数据集使用。

搜集汇总

背景与挑战

背景概述

Sherlock是一个包含超过6000亿数据点的智能手机传感器数据集，采集自50名志愿者的三星Galaxy S5设备，涵盖了多种低权限可监控功能。该数据集特别提供了恶意软件活动的明确时间标签，使其成为机器学习算法测试的理想基准。