Time-series specific Support Vector Classifier (SVC) is an implementation of SVM that works directly on time series, instead of tabular data that could be extracted from the time series. The task of utilizing SVC on SWAN-SF data can be broken down into three main phases:

• Cleaning the time series (e.g., outlier detection, approximating, choosing a proper normalization, etc),

• Tuning the SVC's hyper parameters (kernels and its parameters such as gamma and independent coefficients),

• Evaluation of the model's performance using a proper metric.

Two popular imaging techniques, namely Gramian Angular Field (GAF) and Markov Transition Field (MTF) [Xc], can be employed to convert time series into image-like objects. Since SWAN-SF benchmark dataset consists of multivariate time series, multi-layer images will be produced. After a series of visual verification of the outputs, an state-of-the-art CNN model will be used to handle the classification task. The high-level breakdown of this approach is as follows:

• Transformation of time series to images,

• Applying a CNN model for classification of the produced images,

• Evaluation of the model's performance using a proper metric,

• Understanding the features extracted by CNN, in the 1D space of time series.