Evaluating Different Algorithms for Detecting Change-points in Time Series

Change-point detection is useful in many areas and there are several algorithms developed to cater specific needs such as fast calculations and statistical accuracy. In this report we outline five different algorithms, namely Fused Lasso, Filtered Fused Lasso, Normalized Fused Lasso, Dynamic Programming Approach and our own creation using convolution filtering, the Convolution Method. We evaluate them in the mentioned areas. Convolution Method is the fastest, Dynamic Programming Approach is the most accurate but suffers greatly in speed compared the other algorithms. Normalized and Filtered Fused Lasso performs similar in both speed and statistical accuracy. Although Fused Lasso is as fast as Normalized and Filtered Fused Lasso, it can not detect true change-points to any reliable extent