Upgrade to Pro — share decks privately, control downloads, hide ads and more …

Safe Dining Talk

ahmad0510
February 18, 2016

Safe Dining Talk

An online recommender system to suggest safe eating places in your neighborhood

ahmad0510

February 18, 2016
Tweet

More Decks by ahmad0510

Other Decks in Technology

Transcript

  1. Personal Story What if I could find which restaurants are

    safe and which are not? Mugged @Parking lot ! L Hungry for Pizza! Closest pizza location
  2. I want to eat Pizza at 2+ rated restaurant within

    3 miles of my location Under the hood…
  3. I want to eat Pizza at 2+ rated restaurant within

    3 miles of my location p1 p2 p3 p4 p5 p1 p2 p3 p4 p1 p2 p1 p2 p3 p4 p5 p6 p7 p8 p1 Crime Location Crime Probability
  4. I want to eat Pizza at 2+ rated restaurant within

    3 miles of my location p1 p2 p3 p4 p5 p1 p2 p3 p4 p1 p2 p1 p2 p3 p4 p5 p6 p7 p8 p1 Crime Location Crime Probability 1. pRest 1. pRest 1. pRest 1. pRest pRest Crime Probability at restaurant
  5. I want to eat Pizza at 2+ rated restaurant within

    3 miles of my location p1 p2 p3 p4 p5 p1 p2 p3 p4 p1 p2 p1 p2 p3 p4 p5 p6 p7 p8 p1 Crime Location Crime Probability 1. pRest 1. pRest 1. pRest 1. pRest pRest Crime Probability at restaurant pRest pRest pRest pRest pRest pRest pRest
  6. I want to eat Pizza at 2+ rated restaurant within

    3 miles of my location p1 p2 p3 p4 p5 p1 p2 p3 p4 p1 p2 p1 p2 p3 p4 p5 p6 p7 p8 p1 Crime Location Crime Probability 1. pRest 1. pRest 1. pRest 1. pRest pRest Crime Probability at restaurant pRest pRest pRest pRest pRest pRest pRest 2. Relative Safety Index 2. Relative Safety Index 2. Relative Safety Index 2. Relative Safety Index
  7. Workflow Crime dataset Yelp search API Preprocessing Wrangling Defining classification

    problem Feature Engineering Choosing a model Validation 10-fold cross validation Cross entropy loss measure
  8. ™  Multiclass Classification –  Predict the probability of crime at

    any hour at any given location –  Features: 35 -- Location (Lat., Long. etc.) , Time (dd, wk, mm, yyyy) –  Labels: 24 classes (hour of day) –  Logistic Regression ™  calibrated results ™  Minimum misclassified results ™  Lowest run time –  Cross entropy loss measure = 2.95 Classification
  9. ™ A color coded ranked recommendation list of top 10 safest

    restaurants ™ Location of the restaurants relative to my location in a Google Map display ™ Crime heat map of the neighborhood, if I want to walk to a restaurant Recommendation summary
  10. •  Ahmad Haider •  Georgia Tech •  PhD thesis: “Measurement

    of energy landscapes of biological interactions using Boltzmann sampling” •  Love hiking About Me