Shocking news 70% of all data science projects are failing. But why? We focus ed on this question and wediscovered that it is not the technical side that caused this high failing rate. It’s a combination of challenges but one of the main points is the acceptance of end users and the imagination of future use cases.
RapidMiner is already an excellent tool for making the magic behind data science projects and their solutions transparent. Too many other tools are appearing like black boxes to the end user, the person who should take advantage of the project outcome. This transparency has proven to be a key requirement to gain their acceptance and hence for project success. Unfortunately, this is still not enough in many situations: Machine learning is not widely understood and end users have a hard time to imagine how a solution could look like at the beginning of the project. Misunderstandings when specifying the requirements are the norm, which can be costly or prevent realization of potentials.
So often enough there are good machine learning results, but because of a lack of understanding the results are not used adequately. In conclusion, the end user needs a clear conception of the deployment interface and functions in an early stage of the process.
Where can Web Apps be helpful?
This is where the WebAppBuilder Extension comes into play. It adds the deployment to the agile way of RapidMiner. Effortlessly the data scientist can create the end users’ interface for interactive deployment very early in the project, right inside and with his preferred tool RapidMiner. As soon as there is the very first result, it can be deployed as a web app and shown to the end user. With this tangible and concrete app, they can now imagine how a final solution could look like and the possibilities of machine learning sink in. On that basis they can specify requirements for their daily use, harnessing the power of machine learning. They will start making the project to their project and by that guaranteeing long term success.
At this stage, it makes no sense to spend money on third-party support for example an external service provider that is developing an interface. Even an internal colleague that is not involved in the project who could do this work would simply use too many resources, because many changes will come with going further in the data science process, depending on findings and new ideas or requirements. The responsible data scientist would have to explain all changes to the person who does the deployment and to the end user as well. In real life that could be even more than two people who would be involved.
With all these iterations of explaining, understanding and integrating deployment this concept would be very inefficient and non-profitable. That is why we wanted to enable the data scientist to use the RapidMiner platform directly for deployment in a multifunctional way, for prototyping the deployment, and even substitute other deployment tools.
No need to mention that this makes the process much leaner and efficient even if you would have to use another tool for any reason at the end of the project. It is a lot easier to explain all functions when you can already show a working product.
How to increase your project success?
At the beginning of data science projects in the area of machine learning, end users and domain experts have no actual understanding of what machine learning is. With machine learning being a completely new approach to learning, most people have a hard time embracing it. Letting go of the idea that one has to understand a problem oneself to solve it, is a hard task. Getting a concrete idea of what machine learning really is, helps a lot. If at the same time a new way of getting involved is opened, acceptance is nearly guaranteed.
With the end user being part of the development phase, all functions are already known and the new tool can go live without a long learning phase of its users. This is where end user and data scientist work hand in hand with the same goal in mind.
Additionally, it is often not clear what the computer finds or if it even finds some pattern in the data to help the company. Unfortunately, there is no free lunch in data Science.
But when knowledge keeper and data scientist work together and learn from each other the magic can happen. For this, we invented our Standard Process for Establishing Data Science (SPEDS) which addresses many hitches in the whole change process of organizations. The WebAppBuilder is a perfect aid to support this approach to increase the end user involvement.
As the app can utilize the RapidMiner Servers’ user management and single sign on integration, it is automatically fully enterprise ready. Once a project has reached an acceptable performance level from the data science perspective, an app, developed in parallel, can be put into use at once. This drastically reduces the time to realize benefit with the results and avoids the possibly costly dependency on any extra resources from other departments or suppliers.
Use cases for the WebAppBuilder extension:
Mr Schmidt is the production manager of a paper factory that produces card boards boxes. He wants to know how he can increase production speed and thereby ensure that paper humidity remains good. In the past this experiment more often went wrong and so he engages a data scientist who uses history data to simulate how paper humidity evolves when production parameters are changed. The operator can use this simulation to check the effect of changing parameters before they really will be applied on the machine.
The simulation includes a live monitor where the current production parameters and also the prediction for paper humidity is shown. This live monitor will update automatically every 5 seconds. In the setting parameters the horizon for the prediction and the period of historical data can be adjusted. There are control panels, to simulate what would happen if the operator changes some production parameters. If the operator is satisfied with the prediction of the humidity, he can send the new adjustment into the production. There is also a traffic light to show the paper humidity at a glance. If the light is red for a humidity outside the limit values, the operator will get a notification if he tries to send the adjusted production parameters to the machine.
The operator of a shopping app wants to give sellers the ability to use a BI dashboard to analyze the anonymized data of users made in the shopping app. Sellers thus have the opportunity to better set up their assortment and to discover gaps in the market. For this purpose, the dashboard offers a ranking, which shows which are the most common search terms within the product categories offered by the seller. In addition, it shows how many of these searches ultimately lead to a successful purchase. It will also be checked whether this purchase was made on a seller’s product or on a competitor’s product. This way, sellers can easily see for which search terms the seller is already well positioned, for which searches there is actually a lot of demand, which demand is covered by competitors, etc.
By quickly transferring the analysis results to a web app, a prototype can be developed directly from Rapid Miner. This allows individual users of the shopping app to directly test the functions of the dashboard and the feedback can be immediately incorporated into the further development.
Features of the WebAppBuilder:
Structure and layout
- Multipage Apps with navigation bar
- Modular design using containers and components
- Component arrangement via
- Column-based layout
- Accordion Layout
- About Tabs
- Scatter chart
- Line chart
- Pie chart
- Bar chart
- Bubble charts
- Editable tables
- Forms for data entry and correction
- Frames for the integration of other web resources
- input fields
- Text fields
- Radio buttons
- Drop Down Menus
- Date selection
- File Upload
- Are bound to control components
- Execution of any RM process
- Influencing variables
- Custom Java Script
- Navigation in the app
You can downlod the extension here.
You can buy the license here.