When working on a machine learning project, choosing the right error or evaluation metric is critical. This is a measure of how well your model performs at the task you built it for, and choosing the correct metric for the model is a critical task for any machine learning engineer or data scientist. Log Loss is a common metric for imbalance multiclass problems.

For Zindi competitions, we choose the evaluation metric for each competition based on what we want the model to achieve. Understanding each metric and the type of model you use each for is one of the first steps towards mastery of machine learning techniques.

Log Loss (logarithmic loss) also known as cross-entropy loss is a widely used evaluation metric in the world of machine learning. It is suitable for probabilistic models such as logistic regression or neural networks. The output of such a model is interpreted as the probability of correctly predicting the class label given a list of features. Log loss measures the performance of a classification model by quantifying the difference between predicted probabilities and actual outcomes. Specifically, it assesses the fit of a model's predicted probabilities to the true class labels. In binary classification, the goal is to predict one of two discrete class labels {0, 1} and the corresponding loss function is measured as:

When the predicted probability is high, that is close to the actual class, the log loss is low whereas when the predicted probability for the actual class is small, the loss is high. The beauty of log loss lies in its ability to penalise confident misclassifications heavily.

For a multilabel classification task consisting of C classes, the log loss is given by:

Log loss takes into account the distribution of class labels, making it less susceptible to skewed class proportions than accuracy. This is crucial when dealing with imbalanced datasets where one class significantly outnumbers the other.

With this knowledge, you should be well equipped to use log loss for your next machine learning project.

Why don't you test out your knowledge on one of our knowledge competitions that uses log loss as its evaluation metric ? We suggest the Swahili News Classification.