Hi Team,

Hope you are doing well and enjoying your weekend.

After completing the iterations methods highlighted with blue in the trailing email, I see a slight precision decrease (from 3.9% to 3.01%) in the new bulk model but ROC score has increased (from 56% to 58%). Although it’s not a substantial increase in performance but it’s still better than our previous W-o-W incremental model which gave us a precision of only 0.3%.

@Mayank Thanks for the suggestions on updating Bulk model for better performance.

I am attaching the detailed summaries of all the iterations that I ran today.

Please feel free to reach me in case of any issues.

Thanks and Regards,

Aashish Sharma
Data Science Associate
ZS Associates India Pvt. Ltd.
World Trade Center, Tower 3, Kharadi, Pune 411014, Maharashtra, India

www.zs.com

ZS  Impact where it matters.



From: Aashish Sharma
Sent: Friday, November 16, 2018 10:23 PM
To: Prakash Prakash <prakash.prakash@zs.com>; Brandon Bartell <Brandon.Bartell@zs.com>; Arindam Paul <arindam.paul@zs.com>; Srinivas Chilukuri <srinivas.chilukuri@zs.com>; Mayank Tewari <mayank.tewari@zs.com>
Cc: Vaibhav Bansal <Vaibhav.Bansal@zs.com>; Shivika Singla <shivika.singla@zs.com>
Subject: Patient Event Prediction | Next Steps

Hi Team,

Thanks Onshore team for a great discussion which helped to us come up with more incremental updates for our predictive model and pipeline.

Here are the next steps:

1.	Adding not only the marginal features but all features in bulk model.
2.	W-o-W incremental training with top 3000 patients instead of top 10000 patients.
3.	Combining Bulk model with incremental data training (Detailed discussion on Monday).


Also, we had a great brainstorming session with Shankar which we intend to discuss with the team on Monday.

Please revert in case of any issues.

Thanks,

Aashish Sharma
Data Science Associate
ZS Associates India Pvt. Ltd.
World Trade Center, Tower 3, Kharadi, Pune 411014, Maharashtra, India

www.zs.com

ZS  Impact where it matters.