Alli Fellger CREU Blog 2018-2019

The next experiment I wanted to perform with the kNN regressor was to split the epochs into groups based on constant attributes and then choose the k-nearest-neighbors from that subgroup. The groups I decided to look at were Sex (Male/Female), Light Group (Blue/Red), Physical Therapy Group (Stroke/TBI) and Phase (1/2). These results could then be compared to the results obtained from all instances to see which was more accurate.

This week I worked on the kNN regression model I discussed last week. The appeal of this method would be that instead of attempting to create a model to classify a single instance based only on historical data, I could potentially form a model based on all instances across all patients based on their similarity to the target instance. I hoped that this would yield more accurate results, as there would be more data. Once I had the framework to conduct the experiment, there were several sets of variables I needed to test. The first of these was K, the number of neighbours used. There is not a single way by which I felt confident I could optimize this value without simply testing different values and comparing my results. This became quite time consuming around k = 30, where it took multiple hours in order for the program to run. I eventually settled on k = 15 by trial-and-error testing. The second variable was the set of attributes used to find the “closest” ins

My exploration of inactive minutes vs. Dr. Skornyakov's sleep minutes yielded interesting results. For the overwhelming majority of patients, I found correlation between the two features to be above 0.8, indicating a strong relationship. However, three patients' correlations fell significantly lower, one being only 0.18. There was not a significant difference in the correlation between sleep minutes vs inactive minutes and previous/next days' active minutes. In all but three cases, the inactivity minute count was, on average, less than the number of sleep minutes. With this information going forward, my plan is to keep using both features and see how the accuracy of predictions varies for each. As the semester comes to a close, I am going to be working on developing a kNN regression model to predict the number of active minutes for a day. I am going to calculate the accuracy of these predictions using combinations of: sleep minutes (previous night), active minutes (previo

This week I worked on the high activity tracking idea I had from the past few weeks to try and see if there is a relationship between sleep and the number of high activity minutes during the day. My initial approach to this concept was to look at the number of sleep minutes and their correlation with activity minutes the prior day or the following day. As Alexa and I noticed the first time we began comparing these, the data is all over the place. For each grouping (night-day and day-night), some subjects see a positive correlation and some see a negative one. The strength of these correlations was varied as well. Because of the potential difference in the data for the first and second groups during data collection, we'd been looking for some sort of data-driven way to look at sleep that didn't involve the hard cutoffs we'd used to classify sleep initially. So, in the same way that I'd been looking at "sedentary-like" activity, I decided to try measuring &quo