"No one is harder on a talented person than the person themselves" - Linda Wilkinson ; "Trust your guts and don't follow the herd" ; "Validate direction not destination" ;

October 27, 2017

Day #74 - Feature Generation - DateTime and Coordinates

  • Differ Significantly between numeric and categorial features
  • Periodicity - Year, Date, Week
  • Time Slice - Time past since particular moment (before / after), Time moments in period
  • Difference in Dates (Datetime_feature1 - Datetime_feature2)
  • Special Time period (Medication every 3 days)
  • Sales Predictions (Days since last holiday, Days since weekend, Since last sales campaign)
  • Boolean binary indicating date is holiday or not
  • Sales Context Churn Prediction
  •     (Date Since user registration) - DateDiff
  •     (Date Since last purchase) - DateDiff
  •     (Date Since calling customer service) - DateDiff
  • Periodicity - Day number in week, month, season, year, second, minute, hour
  • Time Slice, Difference between dates
  • This can be used for churn prediction (Likelihood customer will return)
  • In Real Estate Scenario for predictions on Prices
  •     (Distance from School)
  •     (Distance from Airport)
  •     (Flats around particular point)
  • Alternatively distance from maximum expensive flat
  • Centre of clusters and find distances from centre point
  • Aggregated Statistics for surrounding data
Happy Learning and Coding!!!

Post a Comment