Weight loss applications are getting popular. Those apps basically track your lifestyle- calorie budget, track your food, and exercise. Of course, they will also offer you some plans and suggest some low-calorie food. The more interesting thing is they can forecast when you can reach your goal if you use this app, based on your past lifestyle — occupation, gender, your current weight, your historical disease, etc. There are so many interesting weight loss apps, such as MyFitnessPal, Lose It!, MyFitnessPal, Fitbit, and Noom.
For example, the weight management app Lose it which is built by Microsoft. This is a pretty cool application because it can be used not only on Windows phones but also support for the Microsoft Band. It, not just an application, it is more about your life assistant.
There is another weight loss application which is called Noom. This app is an award-winning weight-loss program designed by psychologists & scientifically proven to create real, sustainable results.
When I registered as a new user, there were a lot of specific questions that I need to answer in order to obtain my bio-information and health metrics. So that Noom can calculate how long you might need to achieve your goal based on the backend algorithms. Then after you purchase their plan, it can track your health metrics and recommend you low- calorie food and work out to maintain your activity level.
How is Data Science related to Weight Apps?
Insights
Those dynamic data from users could be structured or unstructured. Data scientists can explore various insights from those datasets. Insights are very vital and helpful for decision makers to make strategic decisions.
For example, some users have the same/similar demographic information/data. However, their progress might be different. Some of them lose weight very fast, but some users’ weight rarely changes. According to users' basic information and daily dynamic data, data scientists could figure out more actionable insights.
Clustering Analysis
The efficiency of weight loss progress is different between individuals. Clustering analysis is also a good algorithm to analyze user’s data. For example, we want to know what is the relationship between weight reduction rate and age. As the image shows below, it seems that the older the user is, the lower the user’s weight reduction rate. However, it is better and convenient to use a clustering analysis to reduce the time cost.
Reinforcement Learning
Besides, Reinforcement Learning also can be used in weight loss apps or health care apps. It basically solves a wide range of new tasks — how to learn intelligent behavior in a complex dynamic environment.
For example, under the policy gradients, it can track users eating and work out habits and warning them about some food that they should not eat, and send them suggested food and workout plan. Then we can obtain users' daily/weekly/monthly health metrics. Under the policy network, we can collect a bunch of ‘successful data’ which is from those users who successfully lost weight and reached their goal. We can have those ‘successful data’ go throughout the network, select random actions, then feed them into the engine to create an optimized model for weight loss apps.
Honestly speaking, people will “cheating” in order to get rewards. Some apps will offer you some rewards if you finish today’s task. It is going to be a big problem — how to get accurate data and how to ‘discipline’ users. In order to figure out what kind of factors/elements or combinations of those factors that can help users lose weight, we can consider “Spare Reward Setting” — instead of getting a reward for daily/weekly/monthly, we can get some successful final goal for an entire plan. Besides, the rewards shaping is a function that needs to guide your policy to some desired behaviors. So wen can give users a reward and the end of the plan and it managed to avoid some behaviors that might not help users to lose weight.
