48
Figure 31: accuracy scores of all models and classification scenarios for case B
It is important to emphasize that these are the final scores that the models obtained, as they
trended towards that value and stabalized there afterwards, with no more fine-tuning
happening, and those are the values that are the closest between the training and test sets.
5.1. Interpreting the Classification Reports
The classification reports, found in Annex C: list of all the classification reports, contain all the
information extracted from the model analysis, for the parameters that showed optimized
results. The report displays four columns of information, “precision”, “recall”, “f1 score” and
“support”. Since most of the models work in a one-against-all way (𝑦 = 1 -positive- if it is the
class we are looking for, 𝑦 = 0 -negative- if it is any other class), there are four possible
outcomes of the algorithm calculation:
• True positive (TP): the entry was positive, and the model predicted positive.
• False positive (FP): the entry was negative, and the model predicted positive.
• False negative (FN): the entry was positive, and the model predicted negative.
• True negative (TN): the entry was negative, and the model predicted negative.
These four percentages that make up the model’s performance for each label, will be used to
create the three metrics of performance for the classification report [31][32][33].
Precision (
) is the measure of how accurate the model’s positive predictions are, how
much it’s able to avoid wrongly labelling something as positive: