I have monthly market share data for N stores and each store will have its own market share trend i.e.
- Trend A : Continuously increasing
- Trend B : Continuously decreasing
- Trend C : Increasing for some months and then decreasing
- Trend D : Decreasing for some months and then increasing
What I need to do is cluster the N stores based on the trend of the market share i.e. all stores having Trend A in one cluster, stores having trend B in another and so on.
Should i treat this as a time series clustering problem or is there some kind of distance/correlation measure which can be used to cluster this data.
What could be the best statistical methods to achieve this result?
I recommend the k-shape clustering method for time series:
Article k-Shape: Efficient and Accurate Clustering of Time Series
You can also find its implementation in the tslearn Python package:
https://tslearn.readthedocs.io/en/stable/gen_modules/clustering/tslearn.clustering.KShape.html
I presume you have the market share data for the N stores over time - the same time periods? What is your objective?
Wayne Fischer yes the market share data for N stores is for same time period. The objective is to segregate these stores based on their trends.
Yes - understood that. But once you've segregated the N stores into Trend groups, what will you do with those results? This could help in discerning most appropriate / useful clustering method.
Wayne Fischer Well the ultimate aim would be to define some competitor/pricing strategy for each of the clusters based on their market share trend.
- Badges
- Science topic
- Similar topics
- Statistics
- Time Series