Using Location Data to Predict the Outcome of Emergency Calls Crime in San Francisco Using Location Data to Predict the Outcome of Emergency Calls John Van Gilder
Caffe Struggles Clearly not designed for use with OSX 10.12 or MATLAB 2015a Designed for use with images, not numeric data Caffe group on GitHub often unhelpful www.xkcd.com/979
Methods Formatted Data: Formatted Network Feature Vector: [Day of the week, Latitude, Longitude] Label: {0,1} Formatted Network Four hidden layers: 10-2-10-2 Gave the best accuracy, ranged from ~40-65% depending on these numbers 150000 iterations
Results Previous groups1 have used naïve Bayes classifier to attempt to classify type of crime as well as specific outcome given location and day of the week ~25% accuracy I simply attempted to classify general outcome given location and day of the week ~65.25% accuracy Nearly identical results whether day of the week was included or not [1] Ang, Wang, Chyou at UCSD: http://cseweb.ucsd.edu/~jmcauley/cse255/reports/fa15/037.pdf