Talks by datascientists

'Hello data scientists in Cape Town! Happy new year. Time for the first 2017 event. This time the event is kindly hosted at Allan Gray. They will also supply us snacks and drinks.

Attendees can park in the Clock Tower parking at the waterfront (the entrance is right next to the office) and have their parking validated at reception. The bar will be open from 5:45 so any early arrivals can enjoy a beer before the Meetup starts.

We will have 2 speakers.

First speaker is Charles Copley from Jumo World and he will speak about 'large social network analysis using Amazon Web services and scalable cloud computing'

abstract: "In many areas it is too expensive to provide traditional financial services: Jumo aims to use inexpensive mobile phone technologies to bridge this gap and promote financial inclusion, connecting mobile phone users with financial services.

We use social network analysis and graph techniques to improve the way in which we provide our services to people. These techniques allow us to reduce costs and pass these benefits onto users."

https://www.linkedin.com/in/charlescopley

Second speaker is Tristan Bergh from Brandseye.

abstract: "Hospitals can improve their financials by reducing their costs without compromising care. To reduce costs, I explored an In-patient dataset with the aim of predicting the length of stay in bed for a group of elderly patients. Given their initial diagnosis and some other variables, a useful model was created that predicted the likelihood of a patient staying in bed 2 days or longer. The model was intended for use at admissions so that operational staff could more effectively schedule care for patients likely to require more bed days. I discuss the model performance metrics and possible improvements to the model using mortality data."

Talk Outline:
Intro: brief outline of the data environment and the nature of the dataset - diagnosis codes, consultant code, demographics.Methodology: some of the R techniques used, removal of factors that possibly degraded model performance; training and test data setup. Results: Model output and performance metrics. I will talk to the AUC, running the model against different bed day thresholds and accuracy. Discussion: I discuss the intended deployment for the model in operational processing, where I required input from hospital staff regarding implementation and whether or not the model could have been used in a live setting. Mortality data and further model improvements.
Q & A

https://www.linkedin.com/in/tristanbergh

Hope to see you at Allan Gray!