Evaluation

The F₁-score measure will be use for evaluation in this competition. F₁-score considers the average precision_µ and average recall_µ of the test as shown in (1), where average precision_µ computes in average, of all activities and falls, of the number of true positives over the sum of true and false positives; and average recall_µ computes in average, of all activities and falls, of the number of true positives over the sum of true positives and false negatives. The greater and close to 1, the better the metric.

Submission File

Results should be submitted in a CSV file containing two columns: (1) timestamp and (2) class. Please, use headlines "timestamp" and "class" as default. For each prediction (row in CSV file), it should contains a timestamp (one-second minimum resolution) and the class predicted.

First, locate the results from Subject 1, Trial 1, then Subject 1, Trial 2, and then Subject 1, Trial 3. Repeat the same pattern for the remaining subjects.

Notice that the testing dataset will contain several raw data in just one second. For evaluation, we will compute the most frequent class during one-second. For example, if the testing dataset contains the following raw data (e.g. timestamp and class columns are just shown here):

Then, our baseline will be:

The first target value was obtained by applying the mode-operator over the values {4,4,4,4,9} in the range between 2018-07-04T12:04:17.738369 to 2018-07-04T12:04:17.945423 (equivalent to second 17).

The second target value was obtained by applying the mode-operator over the values {9,9,9,9,9,9} in the range between 2018-07-04T12:04:18.007088 to 2018-07-04T12:04:18.198098 (equivalent to second 18).

So, your file would look like: