Thought Process

The first thing I thought about when I read the description for this challenge was: who are these trick-or-treaters we’re trying to impress? What are their profiles and their tastes? Put another way - it's not about what the readers of the report like, it's about what their potential "customers" (a.k.a. trick-or-treaters) will want. I want the readers to visualize who they’re going to be treating and stock up on treats accordingly. We’re here to listen to the voice of the candy enthusiasts. We won’t try to educate them or propose healthier options because nobody ever won a popularity contest this way. Also, we all know that there’s a price for popularity - so we’re not going to consider the price as a factor when choosing the final top 3 treats. I considered 2 possible audiences: the busy shopper and the analyst. For the busy shopper - I tried to keep the exploration opportunities for the report user to a minimum. I'm not asking the reader about what ingredients they like in their candy because we already have data about what is popular and what is not. I used interactivity only to allow the user to generate their shopping list and explore more options in a guided and restricted manner. For the candy analyst - I included the Python script that I used to explore, analyze, and manipulate the data.

Methodology

I used K-means clustering to group the treats into categories with similar characteristics. Then I chose the "winner" from each group as one of the final 3 treats - the winner being the treat with the highest win % per group. This way we ensure the variety of the treats - each treat represents a well-defined group (different from the others) that satisfies specific tastes. I then went further and based on these groups I created "trick-or-treater" profiles - to help the reader understand who they're buying the treats for. After some testing, I decided to use only the ingredients to set up the categories. Interestingly, in almost all of my clustering tests (with various combinations) the final winners were the same, but the composition of the groups differed slightly. One of the main reasons I used only ingredients for clustering was because we're interested in 'taste'. Although shape and texture can also influence the delightfulness of a treat, they are not essential. As a subject matter expert in the candy field (someone had to do it), I also accounted for the fact that ingredients can cause allergies or can be simply disliked, so I found it more useful to form groups around that. As a side note, one interesting thing that I tried for the first time while exploring the data was the UpSet plot. While I didn't find it necessary to include it in the final PowerBI report, I included it in the Jupyter Notebook which can be found here: https://nbviewer.org/github/IrinaCo/Maven-Halloween-Challenge/blob/main/Maven Halloween Candy Challenge_final.ipynb The UpSet plot is a data visualization method used for showing set data with more than three intersecting sets. Here we have several ingredients intersecting and forming different treats, so I found it useful in helping me understand the data at a glance. Here's a snapshot of it below. I just found it neat! ![undefined](https://mavenanalyticsio-upload-bucket-prod.s3.us-west-2.amazonaws.com/191775399/projects/4cf9a22e-52f4-4fea-a7f5-fc9be05c4b1c.png)

Design

This is my first and probably only time venturing to the dark (dashboard) side. It just seemed appropriate for this very specific case. I designed all the backgrounds in Canva. My goal was to put the winners at the center and to keep it simple. On the first page of the report, you can see: - The winners - the most important part of the report - A few basic facts about each of the 3 categories from which the winners emerged (ingredients, number of treats per category) - The trick-or-treater profiles - if you hover over each pumpkin placed next to a trick-or-treater, you will see a short description - Your shopping list - the winning treats have different win percentages, so we can expect to need different quantities of each treat. They are calculated automatically in this section of the report. - Busy shopper FAQ - which leads the reader to more options, or the Python code The second page includes more options. However, I strictly limited these options, to ensure that the final choice is acceptable, with a win percent of at least 50%. Here are the selections that the user can make: - Choose between treats with less sugar, budget-friendly alternatives, or second-best options - Adjust the win % - starting at 50% (minimum acceptable threshold) up to 59% (the percentage for which we have options available from all 3 categories) - The scatterplot will modify accordingly - the X-axis changes between sugar, price, and win%, The y-axis constant line will move with the adjusted Win %, and the colors of the markers will change as well (colored by cluster if within the desired range, or grayed out otherwise). - The matrices on the right side will change to reflect the changes in the scatterplot.

Resources

- Design: I used Canva to design the backgrounds of the report - Picking complementary colors: https://colorkit.co/ - Photos: https://pixabay.com/. Here are the photos used for: [Lizzy](https://pixabay.com/photos/girl-child-happy-dress-fashion-5846483/) , [Johnny](https://pixabay.com/photos/cap-boy-smile-tomboy-emotions-2923682/), Marcus - [Upset plot info](https://en.wikipedia.org/wiki/UpSet_plot) - Inspiration for the [dynamic scatterplot](https://www.youtube.com/watch?v=W_aAWZRgA8s) on the second page, from the How To Power BI Youtube channel

The Busy Shopper's Guide to Halloween Popularity

Maven Halloween Challenge Dashboard

Additional project images

More projects by Irina