When a dataset of asphalt pavement type is first used in a paper published in 2018, MiniRocket has not been introduced yet. So, in this video, we will apply MiniRocket, an algorithm to transform time series, to that dataset and see if there is an improvement in accuracy.

The input is a univariate time series, but the length of each observation varies. By the total classes, this is a multiclass classification problem since we have three labels.

The link to the original paper: https://www.sciencedirect.com/science...

The link to get the data: https://timeseriesclassification.com/...

Source code: https://github.com/stephanielees/Asph...

00:00 Intro
02:01 Why MiniRocket?
03:15 MiniRocket (kernels hyperparameters)
22:37 MiniRocket optimization

Application with Python:
32:31 The data
37:28 Applying MiniRocket
39:07 Train a Gradient Boosting classifier and predict

[UPDATE]: I think I summarize the original paper slightly incorrectly 🙏🙏🙏 I said in the video that the paper wants to try to increase the classification accuracy. The better summary might be something like, "the paper tries to improve the prediction of asphalt condition by using accelerometer so that the pavement monitoring can be done automatically."

#timeseries #classification #timeseriesclassification #datascience #machinelearning #pythonprogramming