Meet Huawei Prediction
Hello again everyone,
Have you looked at Prediction, one of AppGallery Connect’s newest services? Let’s take a look at it together in this blog post. Prediction uses your users’ data on the Huawei Analytics Kit to predict their behavior. With a machine learning-based structure, Prediction plays an important role in increasing your interaction with your user masses.
Prediction now supports Android, iOS, and Web platforms. You can use it for free. You can perform 4 different types of Prediction operations: Churn, Payment, Return, Customization.
Churn: If you want to take action on keeping your users active and uninstall the app, you can consider the risk of churn.
Payment: If you have purchasing features in your application, you can consider the likelihood that your users will pay in the future, and you can perceive their behavior.
Return: The Prediction service helps you predict return users and achieve a virtuous revenue gaining process.
Customization: If you need a special prediction task, you can create it based on Huawei Analytics industry features and your needs.
That’s all. I want to stay on the churn procedure for the rest of the article.
If you want to learn integration HMS check this post. I told you before that the Prediction service is powered by data from the Huawei Analytics Kit. You must also add the Analytics Kit SDK for this.
Once all of this is complete, we need to enable the prediction and analytics kit services over the console.
After Prediction is enabled, if Huawei Analytics Kit doesn’t enable, you must enable it from here.
Prediction needs reports from the Huawei Analytics Kit about users. For this reason, we are also enabling the Analytics Kit.
After enabling them and completing the integration into your application, Prediction creates data on the parameters we have specified for 7 days in the future based on the data of the last 7 days of the application.
After the activations, a screen similar to this will welcome you. Then, you can list the forecast types I mentioned above separately in the table. It is the status field specified for each task that you should pay attention to. If you look at my example, it now appears as “Collecting Data”. This means that I have just made the integration and there are not enough reports to generate predictions.
Prediction has 5 different status:
The churn prediction task uses the active user data in the last two weeks to train the model, which predicts the probability that active users of the app in the last week are lost in the next week.
As stated, time is required to generate the necessary reports for active users at the beginning. If no problem occurs, the status column will be “Successful” and we will be ready to view the active users with the risk of churn.
After being successful, the View button in the right area becomes active and you can access the required report by clicking it.
Some rates and graphs welcome you when you open the report. At the top you can find the percentage of positives and negatives correctly identified.
Below, we can display the users with the possibility of loss in 3 different groups as high, medium and low. Prediction offers these users to us this way. Now we can get in touch with them and keep them in practice. It works compatible with Prediction, A / B Testing and Remote Configuration services. Through these services, we can produce some scenarios and communicate with the determined user groups.
For this, a screen opens where we select the high, medium or low masses we want and click the “Apply” button on it. From here we can choose A / B Testing or Remote Configuration.
After selecting one of the two services on this screen, you will be directed to that service. Your application of A / B Tests or Conditions specific to your users with churn risk can be completed in such a simple way.
For example, in the example here, Remote Configuration creates a condition for the churn risky user group.
Conclusion
In this article, What is Huawei Prediction? What does it provide to application developers? How do we identify users at risk of churn? We tried to find answers to such questions. I hope it was useful.
References
If you want to take a look at sample data, you can take a look at Demo Project.
If you have any questions, don’t forget to visit the Huawei Forum!
