Incremental Approach for Early Time Series Classification

Early classification of time series data aims to classify a time series with high accuracy as early as possible. This paper provides a concise yet comprehensive overview of early classification techniques for time series data, specifically focusing on two widely adopted approaches: the multi-model approach and the shapelet-based approach. By examining the limitations associated with these approaches, this study introduces an innovative incremental approach as an alternative. The incremental approach exhibits the ability to learn and adapt the classification model to new data while retaining existing knowledge, without the need to build new models from scratch for each classification task. In the experiments with time series 'occupancy detection' dataset, time series approach, shapelet approach and incremental approach have been implemented. The experimental results clearly demonstrate that the incremental approach outperforms the other methods in terms of both accuracy and earliness. However, it is important to note that the incremental approach exhibited a higher false positive rate, indicating the presence of misclassifications that warrant further investigation and refinement. This work shows that incremental approach is feasible and efficient for early classification of time series, but also shows room for improvement. Overall, this study contributes to the understanding of early classification techniques for time series data, paving the way for improved decision-making and analysis in various domains reliant on timely and accurate classification of time series data.