DTW-Based Hierarchical Clustering for FMCG Sales Time Series
Andrew Chauzov
Data Scientist
Python
Objective: Aimed to analyze and cluster Consumer Goods sales data to discover distinct patterns and trends.
Analyzed time series data, performing interpolation and adjustments to varying lengths, removing about 10% of the data for quality assurance.
Employed Dynamic Time Warping (DTW) as the matching algorithm to effectively identify similarities in time series data.
Created a NumPy-based algorithm, systematically aggregating at least two series simultaneously. This algorithm emulated the logic of traditional hierarchical clustering in metric calculation and data aggregation.
Outcome: The algorithm proved highly successful and insightful, simultaneously enabling the effortless clustering of over 10k time series.
https://towardsdatascience.com/time-series-hierarchical-clustering-using-dynamic-time-warping-in-python-c8c9edf2fda5