Python Program Project

Brain state cognition and intervention research at the Looscan Elementary School

Data Analysis Report

The purpose of this study includes: 1. Cross-compare learning effects between subjects with different aptitudes, including gender and

age. 2. Examine the effect of mindfulness meditation on learning.

Equipment used for the experiment: 1. 9 Muse headbands and 9 cellphones with the mobile app installed.

Targeted population: 2nd grade students, UHD student volunteers, and teachers.

The procedures should have the framework as follows: Learning procedure used in the experiment:

The learning procedure was carried-out on paper. The sets of word puzzles were given to the kids group and adult group, respectively. 1. The kids group was given 10 words puzzle in themes of feelings and emotions in the pre- and

post-intervention learning sessions, respectively. 2. The adults group was given 20 words puzzle in themes of feelings and emotions in the pre- and

post-intervention learning sessions, respectively.

Pre-experiment activities: 1. Demonstrate how the Muse headband works and teach the students how to use it for guided

meditation.

Pre-intervention activities: 1. Ask for 10 volunteers to participate in the meditation training; and pick one of them for EEG

data collection (In this report, we don’t address the data analysis on EEG data). 2. Ask 9 volunteers to complete the learning procedure. The proctors take the following

measurements: 1) The time each volunteer takes to complete the procedure if the total time is less than 5

minutes. 2) The total number of words found if the time reaches 5 minutes maximum.

Meditation: Ask the 9 volunteers to do meditation using Muse headbands as aids. 1 Give the volunteers extra verbal hints to help them concentrate better. 2 The goal is to make the Muse app display the scene that indicates good meditative state. 3 Take the time each volunteer stay in meditative state, for maximal 3 accumulated minutes.

Post-intervention activities: Ask the same 10 volunteers to repeat the same learning procedure with a different setting;

and use the same Subject to collect EED data. The proctors take the same measurements as in pre-intervention activities.

The above procedure was repeated 4 times and data were collected from 34 subjects.

Research and service learning: Analyze the collected data, involving students in the data mining class in data analysis. 1. Analyze the pre- and post-meditation data, observe the data distribution, try to identify

differences between the pre- and post-meditation states. 2. Build a machine learning model to detect the features that differentiate the pre- and post-

meditation states. 3. Verify the machine learning model using the data.

Data: The dataset has 11 variables.

1. Gender: M or F 2. Age: K (Kid) or A (Adult) 3. Pre-Time: Total puzzle solving time in seconds in the pre-meditation learning session. 4. Pre-Count: Total number of words found when reaching the 5 minutes limit. 5. Active: Number of seconds in “Active” state. 6. Neutral: Number of seconds in “Neutral” state. 7. Calm: Number of seconds in “Calm” state. 8. Recoveries: Number of times the Muse app loses connection to the Muse headband. 9. Birds: Number of bird chirps heard. Every time the subject meditate well, there will be a bid

chirp. 10. Post-Time: Total puzzle solving time in seconds in the post-meditation learning session. 11. Post-Counts: Total number of words found when reaching the 5 minutes limit.

Preliminary Data Processing:

Firstly, calculate the learning performance data as follows.

Compute the “learning time per word” by dividing the total number of words found into the total number of seconds used, i.e., for the pre-meditation learning session:

Pre-Learning-Time = Pre-Time / Pre-Count

Compute the post learning time similarly. Then, compute the change between the pre-learning time and the post-learning time by subtracting the post-learning time from the pre-learning time. Add three variables, Pre-Learning-Time, Post-Learning-Time, and Change to the dataset Hypothesis:

Variable Change has negative correlation with variable Calm, which indicates that the more meditation time leads to higher learning rate.

Findings:

1. Adult group shows slightly positive correlation between variables Change and Calm when the value of Calm is less than 150 seconds; and slightly negative correlation when the value of Calm is above 150 seconds (see the scatterplot above)

2. Kids group shows negative correlation between variables Change and Calm when the value of Calm is less than 100 seconds; and positive correlation when the value of Calm is above 100 seconds. Based on 1 and 2, our hypothesis is only true when the kids do a short-term meditation (no more than 100 seconds); and the adults do a longer term meditation (over 150 seconds). In addition, the intervention has much smaller effect on adults than on kids.

3. Applying JRipper rule-based learning, using Gender as the dependent variable, the purpose is to find out whether there is a systematic difference in the learning performance between male and female subjects. Two rules were returned:

(Active >= 4) => Gender=M (15.0/6.0) => Gender=F (19.0/6.0)

The rules read as: If the Active time exceeds 4 seconds, the subject is a male, otherwise, female. These rules correctly predicted 64.7059% of the subjects. These rules indicate that males are more likely to have an active brain state than females.

4. Applying JRipper rule-based learning, using discretized Change variable as the dependent variable, and the Change variable is discretized into 3 values: Decrease, noChange, and Increase.

The purpose is to find out whether there is a systematic difference in the learning performance between male and female subjects. Four rules were returned: (Calm <= 54) and (Gender = M) => change=noChange (3.0/0.0) (preLearningTime >= 150) => change=Increase (6.0/1.0) (postLearningTime <= 8.9) => change=Increase (2.0/0.0) => change=Decrease (23.0/4.0) The rules read as: If the subject is a male and the Calm time is no more than 54 seconds, there is no change in learning rate, otherwise, if the preLearningTime is no less than 150 seconds/word, or the postLearningTime is no greater than 8.9 seconds/word, the learning time per word will increase; in all other cases, the learning time per word will decrease. These rules correctly predicted 85.2941% of the subjects. These rules indicate that males are unlikely to be affected by meditation if the total meditation time is no more than 54 seconds; also, if the subject has a long pre-learning time per word (>= 150 seconds per word), or a short post- learning time per word (<=8.9 seconds/word), the learning time per word will increase; in all other cases, the learning time per word will decrease.

Reflections: 1. Given the small dataset, all findings are subject to further investigation with more experimental

data. 2. Overall, females are more easily to get into calm state than males. 3. Adults’s performance is less affected by meditation than kids. 4. Kids show performance increase after short-term meditation, while adults show performance

increase after long –term meditation. 5. Some UHD students mentioned that the post-intervention word puzzles are harder than pre-

intervention ones. This may create some bias in the collected data. 6. Knowledge-based learning procedure (e.g., word puzzle) may not return prominent

performance change due to meditation. A learning procedure that is directly affected by brain state may be a better setting for such studies. The proposers of this study is planning to use computer game to facilitate the learning process in future studies.