Supervised learning. As the name suggests, the agent learns from a set of labeled input data given by an experienced external supervisor with the desired output. Each scenario is a set of a particular situation and its corresponding action, aiming to find the link between variables to determine a general rule that maps inputs to outputs by extrapolation and regression. Supervised learning has been extensively utilized for channel estimation and spectrum sensing.
Supervised learning. Supervised learning is the machine learning task of creating a dependency between the actual data and training data [89]. Given a set of labelled (known) data set, supervised learning technique attempts to find a function such that it captures the dependency of data as well as minimising the errors in predicting new data. Consid- ering above, one of the popular methods of supervised learning is the curve-fitting technique which we leverage in our work in order to find closed-form expressions in certain cases of interests. Curve Fitting: Curve fitting is a supervised learning technique, where we utilise large sets of (empirical) data in order to obtain explicit analytical expressions. This is particularly useful for drawing insights into the performance of a cellular network, where in general, closed-form expressions do not exist. In this technique, a linear or non-linear curve is placed through the scatter points in the most fitted approach. The fitness of this curve can be measured through different variables including the logarithmic form where root-mean-square-error (the square root of the arithmetic mean of squared values) (RMSE) and variance (the expectation of the squared deviation of a random variable from its mean) as two metrics for error measurements are calculated and minimised.
