Uber ELK tech algorithmn - kNN
“We chose k-nearest neighbours algorithm (KNN), which finds k nearest neighbors (meaning, similar historic trips over a period of time) and then performs a regression on them to create a prediction.”
- Choosing k candidates based on our self-defined similarity function derived from features like: geolocations, time, etc.
- Calculating the weights for each selected candidate based on the similarity function and the weighted average for each response variable as output.
kNN Similarity search:
- Reduce the search space by applying high-level filtering logic, for example, filtering data by city or product ID.
- Perform similarity-based ranking on the reduced data set, and select top-ranked K out of it.
System architecture
Online training model
Engineering Uber Predictions in Real Time with ELK