Machine Learning Employee Turnover in SAP Analytics on Cloud

[vikas_ohri2]

Employee Turonver is  when an employee is replaced with a new employee in any organization.

Can a attrition or turnover Risk Score , be assigned to each employee & use it as indicator for prediction score of any employee leaving ?  what is making this employee, likely to leave & on the contrary what is making this employee stay ?Not only know, How likey is this employee going to leave ? Can we determine when will this employee leave ? and most importantly , What can be done to avoid or prevent it. and what is the impact on organization if this employee leaves ?

With Readily avaliable Artificial Intelligence, Machine & Deep Learning technologies , can computer be used to address these questions ? Importantly, Does data support the answers to these questions ? Can Use of Deep Learning Artificial Intelligence Exercise predict and explain turnover in a way that managers could make better decisions and executives would see desired results ?

A while back i had attempted to get answers to employee turnover questions using predictive analytics library in SAP HANA, using random forest classification techniques.

Thought of getting answers again now with available Machine learning & Deep learning libraries like TensorFlow and with SAP Analytics on Cloud solution for digital boardrooms being available. It is a perfect way to get Employee Turnover Insights and assess its risk impact not only in Human Capital Management Functions but to overall Business.

Shared here are is a point of view, While working on SAP Qualified Package Solution by RenewHR , using sample data  made available by IBM

More details and a working demo are available on vikasohri/EmployeeTurnover

 



 

In essence, indvidualized attrition risk and factors or dominant features contributing towards attrition for each employee can be predicted using deep learning models.

Machine Model Training can be done on HANA or HANA External  Machine Learning Interface , in the example shown it's done externally and the predictive results brought in SAC to explore these with other organizational data elements for detailed explanations.

The top contributing features towards attrition can be further explored in overall organizational context of how well training, compensation, succession, benefits etc. are doing as well as see its impact on core line of business impacting sales, productivity, customer service etc.

[mmcisme1]

I couldn't get the example to work.  But it is probably a security thing.

Interesting blog.  But it begs the question, what will be done with the information.  If it is predicted that an employee is a high risk of leaving.  Why would you want to take the time to train them for the next level?  And a prediction is  just that.  Without knowing what is driving a person - yes we can guess - we don't really know they will leave.  What if the prediction is self-fulfilling?

Just some more things to think about as we design more applications like this for analytics.  It is always a guess. But how accurate the guess is depends on the data integrity.  Even AI can't fix bad data.  (At least not yet.)

[vikas_ohri2]

Sorry to hear it. Browser may not allow running external scripts , if clicked to view the application from within the link provided in the blog writeup above.

Let’s see if this link  Launch Application  on  external page for the shared example works for you. Let me know if this one works or not , i have the application running on another site as well.

Your questions are quite valid and so we need to be careful in concluding that AI can answer or provide correct answer to any problem. You are right prediction (still) depends upon the data.  Classification problems and Deep Learning Models are example of supervised learning based on prior observed results.

So why would you train some one  for next level, if they are flagged high risk of leaving ? – Good question.  It depends. Is  the model showing lack of training  &/or inadequate promotions are contributing factors for high risk ? What is the productivity loss , cost of replacing , recruiting, training new employee for the same position ?

Importantly key aspect still remains how to explain the model ? or can one trust the AI , machine learning  (ML) results . Knowing Model accuracy is generally not enough.

What has been demonstrated in the example is, explanation of the machine learning results , to identify what are the specific factors that are contributing towards employee attrition risk ? These shall vary for each employee.  These still need to be explored further with data driven evidence and with Non ML experts before concluding and putting the model to use.

Thanks for reading

[mmcisme1]

The link looks great!!!

You have some great points.    It does show machine learning results.  I'm sure we will see many more in the future as most of us have just barely started looking into it.