An Evolutionary Approach to Pattern-Based Time Series Segmentation

Online unsupervised state recognition in sensor data

“…A large family of methods for segmenting time series aim to summarize them based on piecewise approximations using different functions (see, e.g., [22]- [24]), where segmentation points are found either by minimizing the model's fitting error or at regular intervals. Chung et al [25] map a predefined set of patterns onto the time series to identify the segments dynamically, while Guo et al [26] directly use a set of approximation functions to select the best adapted models for each segment. Srinivasan et al [12] use the states found in a feedback loop, as contextual information to classify the patterns in the time series in a supervised manner.…”

Forecasting Duration Intervals of Scientific Workflow Activities Based on Time-Series Patterns

“…Time-series patterns can be employed for complex prediction tasks, such as the trend analysis and value estimation in stock market, product sales and weather forecast [2] [5]. The major two tasks in pattern based time-series forecasting are pattern recognition which discovers time-series patterns according to the pattern definition and pattern matching which searches for the similar patterns with given query sequences [7].…”

“…The major two tasks in pattern based time-series forecasting are pattern recognition which discovers time-series patterns according to the pattern definition and pattern matching which searches for the similar patterns with given query sequences [7]. To facilitate complex time-series analysis such as timeseries clustering and fast similarity search, time-series segmentation is often employed to reduce the variances in each segment so that they can be described by simple linear or non-linear models [5] [8]. In general, most segmentation algorithms can be classified into three generic algorithms with some variations, i.e.…”

Automatically Recognizing Stock Patterns Using RPCL Neural Networks

“…Discovering all the finest trends is laborious and somewhat redundant. Therefore, trend segmentation is employed in order to smooth off the relatively unimportant fine trends while keep reflecting the general coarse trends in large [1,4,15]. Segmentation involves first dividing the time series into multiple segments, according to some optimization rules, and then connecting each in a head-to-tail manner.…”