This repository was created for a data science project using biometric data to improve performance at the game of shuffleboard. The project is for an Educational Data Mining course at Teachers College, Columbia University. Fall 2019.
Data was collected using the Sensor Kinetics Pro app on a Samsung Galaxy S10 smartphone that was taped to each player's right forearm while they played shuffleboard at a bar in NYC called Bob's Your Uncle on December 5th, 2019. The 'Multi-Sensor Recorder' function was used, and included the following sensors:
Three Dimensional Sensors:
- Accelerometer
- Gyroscope
- Magnetometer
Derived 3-D Sensors:
- Gravity Sensor
- Linear Acceleration
- Rotation Sensor
Scalar Sensors:
- Proximity Sensor
- Pressure Sensor
Rather than play a 2v2 or 1v1 game as is usual in shuffleboard, each player took turns attempting a maximum score with 4 pucks. Each player would initiate the smartphone sensors, push 4 pucks - one at a time - down the board, then stop the sensors, record the score, and save the sensor data. Then the player would repeat the process two more times (each player had 3 attempts to achieve a maximum score). After the player finished his/her three attempts, the smartphone was transferred to the next player.
Each player's highest score with the cleanest data was uploaded to github. If during an attempt a puck was knocked off the board by another puck, that attempt was ommitted for this analysis and saved for a later one.
After analyzing the data, an intervention was proposed, and each player took three more turns in an attempt to improve their high score.
The purpose of this experiment is to evaluate the differences in sensor data between an expert shuffleboard player and a novice shuffleboard player, then use the data to recommend a strategy for the novice player to improve their high score. There were 5 players: 1 expert, 2 intermediate, and 2 novice.
Our hypothesis was that acceleration along the Y-axis of the accelerometer (the force with which a player pushed the puck down the board) could predict expert vs novice.
We would like to perform the following functions in R Studio:
- Visualize the pre-intervention data
- Run a K-means cluster with centers=2 (expert vs novice) and centers=3 (expert vs intermediate vs novice)
- Determine which sensor made the most significant difference in score (hypothesis suggests the accelerometer)
- Visualize the post-intervention data
- Run a K-means cluster to see if any status changed.