A hybrid algorithm for clustering of time series data based on affinity search technique

Area of Science:

• Data Science
• Machine Learning
• Statistics

Background:

• Conventional clustering algorithms are ill-suited for time series data due to their static nature, leading to accuracy issues.
• Time series data analysis is crucial across diverse fields like finance, business, and medical science.

Purpose of the Study:

• To propose a novel hybrid clustering algorithm for time series data.
• To enhance clustering accuracy and reduce computational complexity for time series analysis.

Main Methods:

• A hybrid approach combining sub-clustering based on temporal similarity and k-Medoids clustering based on shape similarity.
• The algorithm first groups time series into subclusters by time, then merges them using k-Medoids based on shape.

Main Results:

• The proposed hybrid model demonstrates superior accuracy compared to existing conventional and hybrid clustering methods.
• The algorithm achieves accurate shape-based similarity determination with low computational complexity.

Conclusions:

• The novel hybrid clustering algorithm offers a more accurate and efficient solution for time series data analysis.
• This approach effectively addresses the limitations of traditional methods in handling the dynamic nature of time series data.