report.typ

#let raw_title = "The UConn Recreation Center Usage Trends"

// Sets the metadata on the pdf
#set document(
    title: [#raw_title],
    author: ("Emmet Spaeth", "David Matos"),
    description: [Abstract goes here],
    //keywords: ("informative", "keywords"),
    date: auto
)

#set page(
    paper: "us-letter",

    // Numbering the pages
    header: context [
        #smallcaps[#raw_title]
        #h(1fr)
        #counter(page).display(
            "1",
            //both: true,
        )
    ],
    //numbering: "1",
)

#set heading(numbering: "1.1.")
#set par(justify: true)
#set align(left)

#set text(
    font: "Libertinus Serif",
    style: "normal",
    size: 12pt,
)

// Function to create a title
#let title(main_title: none, subtitle: none) = {
    set text(
        size: 16pt,
        weight: "bold",
    )

    set par(
        justify: false 
    )

    set align(center)

    v(2em)

    if subtitle != none {
        block(main_title + ":")
    } else {
        block(main_title)
    }

    block(subtitle)
    v(1em)
}

// Function to create the abstract
#let abstract(body) = {

    set par(justify: false)
  
    set align(center)

    v(1em)
    // White space above the title
    block(
        height: 0.2em,
        ""
    )

    text(weight: "bold", size: 14pt)[_Abstract_]

    block(
        above: 1.2em,
        body
    )
    v(1em)
}

// =========================================
// ||     The start of the article        ||
// =========================================

// The title
#title(
    main_title: [#raw_title],
    subtitle: "A Student Led Statistical Anlysis"
)

// Authors
#grid(
    columns: (1fr, 1fr),
    align(center)[
        Emmet Spaeth \
        University of Connecticut\
        #link("mailto:emmet.spaeth@uconn.edu")
    ],
    align(center)[
        David Matos \
        University of Connecticut\
        #link("mailto:david.matos@uconn.edu")
    ]
)

// Abstract
#abstract([
    The aim of this paper is to determine trends in the UConn rec center usage.
    This paper will analyze the occupancy of the rec center both over the
    course of a day, broken into fifteen minute increments, and over a week,
    comparing each day. Then, we will to explain the trends in the data.
    Notably, the difference in usage between the days of the week and the
    periods of growth throughout a day. Finally, we will provide our
    explanations of these trends, advice to improve the quality of service and
    potential cost saving measures.
])

#outline()
#pagebreak()

= Introduction

//What is the topic and why is it worth studying? – the first major section of
//text in the paper, the Introduction commonly describes the topic under
//investigation, summarizes or discusses relevant prior research (for related
//details, please see the Writing Literature Reviews section of this website),
//identifies unresolved issues that the current research will address, and
//provides an overview of the research that is to be described in greater detail
//in the sections to follow.

When should I go to the rec? This is the driving question of the paper. Many
students have shared the experience of having to change a workout because all
the machines they need are full or having to work in with a complete stranger.
It can be intimidating and uncomfortable for some people to work out when the
rec is at its busiest. Equipped with the knowledge of when the rec is busiest,
many students feel comfortable going to work out for the first time.

The current solution provided by UConn is through the rec center app. It's a
webpage that tells you the how full the rec is as a percent of its total
capacity. The webpage acts like a ping, only telling you the capacity at that
exact moment. It can't tell you what the capacity will look like in a couple
hours when you actually want to go work out. Unless you spend a ton of time
paying attention to occupancy on the app, you have no clue which times or days
have the least people. The app also can't tell you if it's busier than usual or
just a random fluctuation, you just have to go and find out. We set out to get
real data to solve these problems.

= Methods

//What did you do? – a section which details how the research was performed. It
//typically features a description of the participants/subjects that were
//involved, the study design, the materials that were used, and the study
//procedure. If there were multiple experiments, then each experiment may require
//a separate Methods section. A rule of thumb is that the Methods section should
//be sufficiently detailed for another researcher to duplicate your research.
//
== Data collection

The only source of information that is publicly available is the capacity page
in the rec app. Many of UConn's apps are made by embedding Chromium and using
web technologies (HTML, CSS, JavaScript), most likely through the popular
Electron framework. This mean once we found the underlying webpage a simple web
scraper could record the occupancy.

After finding the web
page(https://app.safespace.io/api/display/live-occupancy/86fb9e11) we wrote a
script in python to read the value and record it with a time stamp. The script
was set to run every fifteen minutes with a cron job on the server.

The information on the web page was recorded into a CSV file with the following
format. If the script threw any kind of error while running, it would simply
record "Error,Failed to Pull Data at *_DATETIME_*". With a successful execution
the script would record, in this order, the month, day of the month, day of the
week, time, and the data it read (the occupancy).

For example, here is the first five entries:

```csv
Month,day,weekday,time,occupancy
Oct,10,Fri,18:30:04,408
Oct,10,Fri,18:45:04,450
Oct,10,Fri,19:00:09,469
Oct,10,Fri,19:15:04,471
Oct,10,Fri,19:30:04,486
```

Data was collected starting October tenth and ending December tenth. There was
a brief period without any date due to an unexpected outage. Thanksgiving break
was another period without usable data leading to a total of nineteen days down
during the data collection period.

== Data filtering and cleaning

//Removing failed pulls, timezone shift, day light savings time, in hours,
//weekday vs weekend seperation.
Our data cleaning pipeline is as follows:

1. Read CSV data using the pandas library.
2. Convert the time and date related columns into a single datetime object.
3. Remove entries where the data was missing or invalid.
4. Adjust the timezone to Eastern Standard Time (EST).
5. Handle the transition during day light's savings time.
6. Filter data from outside of the operating hours of the rec.
7. Group data by day of the week, weekday/weekend, or by hour of day.
8. Visualize and average the pre mentioned groups.

= Results

//What did you find? – a section which describes the data that was collected and
//the results of any statistical tests that were performed. It may also be
//prefaced by a description of the analysis procedure that was used. If there
//were multiple experiments, then each experiment may require a separate Results
//section.

The first piece of analysis we did was a graph of the average occupancy for
each day over time, shown in @daily_average_over_time. It was immediately
obvious where the gaps in our data were. However, on first glance our data
seemed to be fairly random. The mean occupancy was always lower than the
median, suggesting outliers may be bringing the mean down or central values
were significantly greater than their surrounding data.

#figure(
    image("dailyAverageOverTime.png"),
    caption: [A graph of the daily average occupancy over time],
) <daily_average_over_time>

Our next step was a breakdown of the average occupancy by day of the week,
shown in @daily_average_by_day_of_week. This graph gave a lot of insight.
First, it told us the most popular day to go to the rec was Wednesdays.
Followed closely by Mondays and Tuesdays. After which, there is a steep drop
off in usage starting on Thursday with each following day seeing less usage
than the last. Sunday is the least popular day to go to the rec.

Looking back at @daily_average_over_time we can actually see this cyclic
pattern. It's important to remember the limited hours at the rec center on the
weekends are probably contributing to the lower daily average. However, the
reduced hours on the weekend doesn't explain the large difference between
Saturday and Sunday.

#figure(
    image("byDayofWeek.png"),
    caption: [A graph of the daily average occupancy by day of the week],
) <daily_average_by_day_of_week>

The most interesting of the data is the graph of occupancy over an average day.
The peak on both weekdays and weekends is around hour 17 (5pm). This is closing
time on the weekend, but on weekdays it means a decrease in total occupancy for
a couple hours until closing time.

There are three main periods of growth during a weekday. One from 6-8, the
second from 11-13, and the last from 15-17. On the weekend the first two hours
of the day 10-12 sees the greatest growth. After that period, growth slows and
some what stagnates, but occupancy doesn't decrease until closing.

#figure(
    image("byHourofDay(Weekday_vs_Weekend).png"),
    caption: [A graph of occupancy over time separated by weekday vs weekend],
) <occupancy_over_time>

To further analyze the flow of people in and out of the rec we graphed the
first order derivative of the occupancy over time graph in
@derivative_of_occupancy_over_time. In @derivative_of_occupancy_over_time we
see the same periods of growth in the first graph.

#figure(
    image("changeInOccupancy.png"),
    caption: [A graph of change in occupancy over time],
) <derivative_of_occupancy_over_time>

= Conclusion
//What is the significance of your results? – the final major section of text in
//the paper. The Discussion commonly features a summary of the results that were
//obtained in the study, describes how those results address the topic under
//investigation and/or the issues that the research was designed to address, and
//may expand upon the implications of those findings. Limitations and directions
//for future research are also commonly addressed.

== Findings

To answer our original question, the best time to go to the rec is early Sunday
morning. Remember, the times with the fewest people are in the early morning 
before the end of the 6-8 growth period and the days nearest the end of the week. 
For people with already established programs or early morning classes, a simple
solution is shifting the start of your program forward a day to two to avoid
the busiest days of the week.

== Limitations
//Problems with data collection, limited analysis (no std dev), don't know the
//number on the app is perfect

 - #context [Losing two weeks of data is a significant portion of the dataset
and may have affected the results.]
 - #context [Students are only required to scan an ID on the way in, not out.
This makes the true occupancy unknown because they don't know exactly how many
people have left.]
 - #context [The occupancy on the website doesn't truly reflect how full the
Rec center feels. For example, a fitness class may inflate the occupancy, but
because classes are limited to studios it won't affect regular occupants.]
 - #context [Fifteen minute increments lead to coarser data, which limited the
quality of the data]

== Future Research
//Check if occupancy goes up with every time a person taps their IDs. Basically,
//sit near the entrance to the rec with a counter and see if it matches up.
//Interview students for a more qualitative approach/information. More data
//points. Compare to commercial gym usage. Look at class schedules and dining
//hall occupancy.

We hope to remedy some of the limitations of this study with future research.
The first of which is checking how accurately the website tracks the number of
people actually in the rec. This would be done with one person, sitting by the
entrance counting people coming in and leaving. Then, comparing what they
counted with what the website said at the same time. Collecting more data would
also be helpful. For example, getting an entire semester of occupancy data
would lead to better analysis.

Another avenue of research is needed into the other factors affecting when
people can go to the rec. The most obvious would be tracking occupancy of
dining halls along side the rec. The stagnations in usage growth occur around
regular meal times, suggesting a significant number of people are eating. It's
also important to consider how the distribution of class schedules affects when
people can work out.

Finally, interviews with students may provide reasoning behind the usage
patterns we have observed. The current research has a lack of qualitative data
which would be solved by interviews or questionnaires. Interviews or
questionnaires may also reach students who don't regularly use the rec center.