Time Series Classification using Motifs and Characteristics Extraction: A Case Study on ECG Databases

Abstract

In the last decade, the interest for temporal data analysis methods has increased significantly in many application areas. One of these areas is the medical field, in which temporal data is in the core of innumerous diagnosis exams. However, only a small portion of all gathered medical data is properly analyzed, in part, due to the lack of appropriate temporal methods and tools. This work presents an alternative approach, based on global characteristics and motifs, to mine medical time series databases using machine learning algorithms. Characteristics are data statistics that present a global summary of the data. Motifs are frequently recurrent subsequences that usually represent interesting local patterns. We use a combination of global characteristics and local motifs to describe the data and feed machine learning algorithms. A case study is performed on three databases of Electrocardiogram exams. Our results show the superior performance of our approach in comparison to the naïve method that provides raw temporal data directly to the learning algorithms. We demonstrate that our approach is more accurate and provides more interpretable models than the method that does not extract features.

Explainable AI for Time Series Classification: A Review, Taxonomy and Research Directions

“…Due to possible computational problems in extracting subsequences of the most suitable length, often approximations such as SAX are performed, as detailed in the following. In [10], Maletzke et al propose to extract motifs, i.e. repeating patterns, and characteristics, i.e.…”

“…For this reason, these approaches need further tweaks to perform better, and abstractions need to be easier to interpret for users. For example, motif-based models do not have state-of-the-art accuracy with small, easy-to-understand decision trees [10]. A challenge with using subsequences is the segmentation of the data.…”

“…We believe that time series should receive the same research attention since they are omnipresent, e.g., in technical systems [8], [9], the medical domain [10], or business applications [11]. Further, due to the tremendous amount of data generated by sensors over time, machine learning models yield superior results at many tasks due to their capability to capture long-as well as short-term patterns in the data [12].…”

“…Further, due to the tremendous amount of data generated by sensors over time, machine learning models yield superior results at many tasks due to their capability to capture long-as well as short-term patterns in the data [12]. Thus, such models can outperform experts in certain time series tasks, enabling their application in various use cases, e.g., in predictive maintenance [8], [13], heartbeat anomaly detection [10], or texture recognition [14]. The research field of XAI for time series classification has become more popular since around 2019, a variety of valuable papers have been published in recent years (see Figure 1).…”