# Gesture Recognition — Data

来源：https://www.kaggle.com/competitions/gesture-recognition/data

本地 Meta Kaggle 快照；不是本次实时页面。

# Dataset Description

Using the phone application, Sensor Logger, accelerometer and gyroscope data was collected from a user for the following gestures(Target classes):
- Walking
- Running
- Waving
- Stretching
- Squatting 
- Jumping  
- Dancing
All data was collected while in the user’s pocket and again, while around their arm.Following the instructions made, accelerometer and gyroscope data were collected from the given gestures 
Note: specifically, the vector data given on a XYZ plane, will be used.

**Key Features and Their Purpose**

Timestamps and Time Elapsed: These columns are essential in understanding the temporal flow of the data. The **time **column records the exact time of measurement, while **seconds_elapsed** provides the duration from a defined start point, likely indicating the specific moment each sensor reading was taken.

Acceleration and Gyroscope Measurements: The AccelerationX, AccelerationY, and AccelerationZ columns record acceleration forces along the three axes. This data can indicate the direction and speed of motion. For example, high values in these columns could correspond to activities like running or jumping. The GyroscopeX, GyroscopeY, and GyroscopeZ values measure angular velocity, helping capture orientation changes during movement, such as rotation or tilting.

Activity Label: This target variable associates each sensor reading with an activity, such as "running" or "stretching." Such labels are essential for supervised machine learning tasks, where models can learn patterns in the sensor data that correspond to each activity. This label could also aid in developing applications to track exercise or detect falls.

## Files

*   **train.csv** - This dataset likely captures motion and orientation data from wearable devices or smartphones, possibly used in activity recognition research. It has approximately 47,455 rows, each representing a moment in time where sensor readings were recorded. Each entry is timestamped and includes acceleration and gyroscope data, measured along three orthogonal axes (X, Y, and Z). The data also includes a label that identifies the activity being performed at that moment, which could be used in models for classifying activities based on sensor data patterns.

*   **test.csv** - The test dataset has 11,864 entries with 9 columns, similar to the training dataset but without the target activity label. Here’s a detailed description:
Each entry in the test dataset represents a moment in time with sensor data readings, which are structured for tasks such as activity recognition. Like the training dataset, it includes timestamps and measurements from accelerometer and gyroscope sensors.

This dataset can be used as input for models trained to recognize patterns and classify activities without requiring predefined labels in this file. It serves as unseen data for testing or validating predictive performance, especially in contexts like fitness monitoring, health assessment, or sports analytics, where sensor data is used to infer physical activities.

The lack of a target label in this dataset suggests its use as an evaluation set, where predictions will be compared against hidden labels for model accuracy assessment.


## Columns

- id: A unique identifier for each record.
- time: A timestamp, possibly in a UNIX format.
- seconds_elapsed: Time elapsed for each measurement in seconds.
- AccelerationX, AccelerationY, AccelerationZ: Acceleration values along the X, Y, and Z axes, respectively.
- GyroscopeX, GyroscopeY, GyroscopeZ: Gyroscopic values along the X, Y, and Z axes, respectively.
- Label : Target(classes)