Rendani Mbuvha
18 October 2017 | 11:45 – 12:45 | Committee 2
Relevant practice area(s):Data Analytics
Suggested audience knowledge level: Foundational and Intermediate
Abstract:
This workshop will provide practical ways for incorporating non-traditional data such as speech, text and images to enhance our understanding of actuarial and financial risks. Participants will work through a short exercise on how to build predictive models on newsfeed data using bag-of-words features(rating factors) in R. Details on how to extract meaningful features using Mel-frequency Cepstral Coefficients (MFCCs) for speech and histogram oriented gradients for images will be discussed. This will be further motivated by a practical short-term insurance example of using convolutional neural networks on dashcam images to detect distracted drivers.