ROAS Case Study
NextBrain is a software that makes it easy for you to work with ML, without needing to have any prior programming knowledge. It allows you to explore your data, train ML models, and get insights into the performance and accuracy of your models, as well as the important features in your data. With NextBrain, you can quickly and easily build regression and classification models and use them to make predictions with confidence.
Let's now go through all these features by building a model from scratch. We will work on Marketing and Sales workspace in this tutorial, and predict Return on Ad Spend (ROAS). You can click on your profile on the top right, and modify industries and add financial services to be able to see Marketing and Sales dataset.
First page you will encounter after clicking on a workspace is the problem description section. You will be given a brief summary about the general context and what each column means.
Next, you can click on Explore data on the top bar menu to observe basic properties of your data. Explore data section mainly consists of three sections: Dataset, data statistics, and variable correlation.
Dataset is shown with datatypes for each column, and how it is distributed across different values. You can also see the number of empty entries for each column.
Data statistics shows the maximum value between linear and non-linear correlation between each two features.
Variable correlation is simple a scatter plot between each two features you can select on x axis and y axis bar on top of the graph.
Now, let's move forward to the Train model section. You have a lot of flexibility to build and train your model with a few clicks on NextBrain. We can start with selecting the ROAS feature on column to predict menu. We can now see various options appearing in the same drop-down menu:
Columns to ignore: You can drop some columns on your dataset either for a faster performance, or you are sure some columns wouldn't considerably contribute to model performance. However, keep in mind that dropping some columns can result in poor metrics results.
Training quality: You have Quick, Performance, and Accurate options to build your model. The longer you train your model, the more chance it will yield better results. The times it takes to build each option are around 1, 5, and 10 minutes, respectively.
Advanced: You can feed synthetic data to your model if you don't have enough or balanced data; choose which metrics your model should prioritize, and arrange train/test split ratio.
After setting features that best fit your needs, you can train your model, and obtain your model in a few minutes! A rough summary about your model will appear on the right side. In this summary section, you can see model accuracy compared with a baseline method and realize which columns have little influence on the model performance.
The model accuracy results for previous iterations will be given below the model accuracy graph. Moreover, you can also observe column importance, visual explainability and model training history on the same section. For more details about those widgets, you may go to Dashboard widgets articles here: https://help.nextbrain.ai/en/category/dashboard-widgets-1tgrcb/.
You can also click on Dashboard for a more detailed report about your model. Don't forget to click on the help icon for more comprehensive explanations about widgets. These are some of the features you can view on the Dashboard section:
Algorithms used: NextBrain uses a combination of different models to yield the best performance, and you can see the ratio of each model used in your final model.
Performance: You can compare the performance between your model and a baseline method. For instance,
Column importance: You can see how each column contribute to the final output.
Advanced statistics: Final metrics results for your model.
Feature importance: A Sankey diagram demonstrating the flow of predictions.
Finally, you can predict new instances after entering values for each features the model is trained on.
Let's now go through all these features by building a model from scratch. We will work on Marketing and Sales workspace in this tutorial, and predict Return on Ad Spend (ROAS). You can click on your profile on the top right, and modify industries and add financial services to be able to see Marketing and Sales dataset.
First page you will encounter after clicking on a workspace is the problem description section. You will be given a brief summary about the general context and what each column means.
Next, you can click on Explore data on the top bar menu to observe basic properties of your data. Explore data section mainly consists of three sections: Dataset, data statistics, and variable correlation.
Dataset is shown with datatypes for each column, and how it is distributed across different values. You can also see the number of empty entries for each column.
Data statistics shows the maximum value between linear and non-linear correlation between each two features.
Variable correlation is simple a scatter plot between each two features you can select on x axis and y axis bar on top of the graph.
Now, let's move forward to the Train model section. You have a lot of flexibility to build and train your model with a few clicks on NextBrain. We can start with selecting the ROAS feature on column to predict menu. We can now see various options appearing in the same drop-down menu:
Columns to ignore: You can drop some columns on your dataset either for a faster performance, or you are sure some columns wouldn't considerably contribute to model performance. However, keep in mind that dropping some columns can result in poor metrics results.
Training quality: You have Quick, Performance, and Accurate options to build your model. The longer you train your model, the more chance it will yield better results. The times it takes to build each option are around 1, 5, and 10 minutes, respectively.
Advanced: You can feed synthetic data to your model if you don't have enough or balanced data; choose which metrics your model should prioritize, and arrange train/test split ratio.
After setting features that best fit your needs, you can train your model, and obtain your model in a few minutes! A rough summary about your model will appear on the right side. In this summary section, you can see model accuracy compared with a baseline method and realize which columns have little influence on the model performance.
The model accuracy results for previous iterations will be given below the model accuracy graph. Moreover, you can also observe column importance, visual explainability and model training history on the same section. For more details about those widgets, you may go to Dashboard widgets articles here: https://help.nextbrain.ai/en/category/dashboard-widgets-1tgrcb/.
You can also click on Dashboard for a more detailed report about your model. Don't forget to click on the help icon for more comprehensive explanations about widgets. These are some of the features you can view on the Dashboard section:
Algorithms used: NextBrain uses a combination of different models to yield the best performance, and you can see the ratio of each model used in your final model.
Performance: You can compare the performance between your model and a baseline method. For instance,
Column importance: You can see how each column contribute to the final output.
Advanced statistics: Final metrics results for your model.
Feature importance: A Sankey diagram demonstrating the flow of predictions.
Finally, you can predict new instances after entering values for each features the model is trained on.
Updated on: 10/02/2023
Thank you!