Mood Planet

UX/UI Design & AI

In my master thesis, I was studying mood trackers as an example of design considerations when artificial intelligence (AI) is applied in the context of wearables designed to help mental well‐being.

I focused on what designers should have to consider the relationship between machine learning and

the user when creating a concept of mood tracking applications based on machine learning models.

I created a clickable prototype of a mood tracking application, which is focused on the understanding of the machine learning model and generating trust for the user. The users can use the prototype to become aware of their mood and reach mental well-being.

Master thesis at FH Potsdam


Understanding users & AI, research, ideation, conception, creation of prototypes



Current market situation:

There are enough applications for tracking mood

by asking, “how are you feeling today”.

Still users have to add their mood in the application.

Users have to add their mood by themselves. It is not easy to do it regularly for users.



They are in their mid-twenties to the end of their twenties



They want to track their mood to see patterns of their mood states



They lack the self-awareness of their mood.

“Knowing something does not mean that you are doing it”

Users do not track their mood by themselves

Users can still be aware of their mood

How might we...

enable users to do nothing

leading them to the very best something?


To solve the user’s problem, I was inspired by Winne the Pooh, quoted by one of the people I interviewed. Like Winne the Pooh said, users want that doing nothing leads to the very best something for them.

What if:

Artificial Intelligence detects users’ mood by data, so users can clearly know their mood states.


I got an idea from research that biometric data could help to identify and relieve stress. The combination of comfortable and energetic feelings describes users mood states. And comfortable and energetic feeling strongly correlate with data on heart rate, activity, and GPS. Smart watches can collect those data. With machine learning, it is possible to detect a user's mood by data.

Mood data visualization
as four different mood states

I visualise comfortable and energetic feelings in two dimensions. The combination of both feelings describes users’ mood states. Using the combination of two dimensions helps the users to be aware of their mood, can control AI's detection. Also, it makes users understand their own data.


Furthermore, users do not want to see a negative expression of mood states by AI, as users reported during user testing. Thus, instead of an expression of mood states, the possible situation for each mood state is described.

Considerations for

AI model and privacy issues


Most users are worried about privacy. Using appropriate models (Gradient Boosting, Federated learning) can be an answer to the issue. With proper models, users' data is only stored on their phone or watch, and they can delete it.


As design to helping for users' understanding, users have a chance to read about data usage and algorithms of the Mood Planet application.

Concept of mood tracking application


According to the research, mood tracking applications have four stages: preparation, collection, reflection, and action. I created the structure of the application at this stage. I also thought about the guidelines for AI design that should be considered at each stage.



To prove my idea, I made a paper prototype for mobile and Apple Watch for three different user scenarios. After user testing with the paper prototype, I created a digital prototype with the Adobe XD:

User scenario: Downloading the Mood Planet (Preparation Stage)

Users get an explanation about the Mood Planet app: visual language shows mood states and how the Mood Planet application can detect and itself track users’ mood states.


Most users are worried about privacy, which is a typical problem of AI. So, users have a chance to read about data usage and algorithms of the Mood Planet application.

User scenario: Giving feedback to the Mood Planet (Collection Stage)

In the beginning, the app needs regular feedback from users. When users get a daily notification, users can give feedback with just three clicks through the Apple Watch.

User scenario: Overview in the Mood Planet app 1 (Reflection Stage)

‘History’ shows the weekly mood states and each day’s main mood state. It also contains information about where the person had those mood states and health data.

When a user clicks on the small icon right up on the main mood states, it shows mood states tracked per hour by the AI and other collected data.

‘History’ also contains a calendar showing a monthly overview of mood states. Through the mood states visualization, users can

recognize not only their mood state but also their level of energy and level of comfort.


If the AI detects the user’s mood incorrectly, the user can recognize which part of the two dimensions the machines detect incorrectly.

‘Highlight’ shows the connection between the user’s mood states, activities, and location compared to the last week, month, and year. Thus, users can recognize how their mood states changed.

User scenario: Overview in the Mood Planet app 2 (Action Stage)

Depending on the connection between the user’s mood states and activities on specific locations in the past, AI forecasts how the user can feel in the future.

It suggests activities and locations that have a good effect on the user’s mood.

‘Learning from neighbours’ shows mood states from local neighbours. These maps are generated by the data, which users anonymously donate to improve the service.


Through this information, the user can get some inspiration or feel less alone having a specific mood often.

Outcome & Impact


Detected a user’s mood states

Learning ‘new knowledge’ = Broaden knowledge



Think about own mood states

Gives feedback about the detected mood states

The designer should consider both journeys, the machine’s journey, and the user’s journey of the machine learning model. Worrying about privacy, users are taking a conservative stance on machine learning that uses their personal data. As machine learning needs time to learn about individual users, the benefits of them don’t immediately appear.

Working process

Design Thinking session

At the beginning of my research, I did a Design Thinking session with my community people to get inspiration and clarify target users and their point of view. At the end of the meeting, we thought about many business ideas that can help target users.

Co-creation with users

I did co-creation with my target user group in order to find suitable mood data visualizations for my target users. Before diving deep into the research of mood data visualization, I needed to know the user’s perspective first. Through co-creation I learned which direction I have to proceed for further research and want to get some ideas directly from my target users. Furthermore, I wanted to test if mood data visualization is helpful to my target users or not.

Empathy map

Point Of View

As-is Scenario


Assumptions & Questions

Identify Data

Help AI understand

Confusion matrix

Really/User reaction

Machine output/prediction

To-be Scenario

Low fidelity Wireframe

If you would like to know more about the master thesis, I can send you my master thesis as a PDF.







Make your Browser window wider


Please use your computer like normal-sized species.
In the case of a tablet, make landscape mode,
or use your mobile device.