Sunday, 16 April 2017

Developers Conference 2017 - Improving Machine Learning Models

The third edition of the developers conference of Mauritius was held between the 30th of March and 1st of April 2017. The conference ran on 3 days with 3 tracks in parallel and had about 330+ attendees and 50+ speakers.

My third session at the DevCon was about the techniques that can be used to improve the performance of machine learning models.
The session was just after the introduction to Data Science that was presented by Arwin and myself.



I started with a summary of the data science process and presented the data set that shall be used in the demo.
Next, I explained the steps of data cleansing and made a live demo of same. The attendees were quite surprised that the model got an increase in accuracy of 10% just by doing the cleansing properly.
I then proceeded with more advanced techniques such as Feature Selection, Feature Scaling, Parameter Tuning and Cross Validation.




I noticed that there were less people in this session, maybe about 25 people who were mostly experienced in Data Science. The session ended with some questions from the audience, again regarding the algorithms and the ability to deploy the models.
I also got questions from enterprise people interested to apply and use the Advanced Analytics features of Microsoft Azure. Looking forward to hear more about it soon!

Please find below the slides used during the session:




Cheers!
Chervine

No comments:

Post a Comment