Toronto Neighborhood Segmentation
Description on the problem
I have read and heard from friends about advantages of migration to Canada and especially to the city of Toronto, My status as family father and my business as an independant consultant (banks and finance) involves a lot of conditions before taking a big move like moving to a new address. More generally this analysis will benefit to many famillies having the same objectives.
Objective
My objective is to evaluate the the opportunity of living in Toronto neighbourhoods based on 4 Criterias
- Proximity to green spots. (family confort)
- Proximity to banks. (workplaces)
- Proximity to beach. (clean air)
target audience
This analysis will be useful for families considering a move or a migration to the city of Toronto. The choosen criterias can be changed to adapt to specific needs.
Data collection and usage
Data will be mainly prepared as follows :
- Boroughs and neighborhouds names will be collected by scrapping wikipedia ( Toronto's addresses page in particular).
- Geolocalisation will be done using geocoder library and opencage API
- Venues collection will be done using foursquare API.
Once done the data will be grouped by venue's type concentration for each neighborhood. Finally the neighborhoods will be ranked depending on best scores (based on target venue's types concentration). For instance Victoria park neighborhood is a perfect candidate for beeing surrounded by parks, near to important economic areas and not too far from the beach.
Methodology
We are aiming for a cluster of target neignborhoods that share nearby features by using unsupervised classification (using KMeans). Then we will score each cluster with a weighted map of features (venue types) that will favor the closest cluster to our preference.
the parameters for this study are :
- The weighted list of venue types that are of interest for a given case. (in my case proximity to work and quality of life)
- The number of clusters to group neighbourhoods (More clusters means that we are more selective and want reduced choices)
The winner is the cluster composed by :
- Berczy Park
- Garden District, Ryerson
- Queen's Park, Ontario Provincial Government
- The Beaches
Discussion
Applying bias (weighted map) can be done before segmentation but it would affect the unsupervised clustering, we choose to apply it after clustering to affect only scoring.
Normally we should add cost of life parameters but it's undirectly included in the type of venues (the more popular the less costly).
Conclusion
We have done a parametrized study to compare neughboorhoods of the city of Toronto, we focused on a specific case "Fitness for Migration given family preferences". We naturally got a short list of condidates, next step will be manual and will take into consideration non formal and human reviewed criteria.
