Aspects for machine learning-based link adaptation are described. For example, an apparatus can determine k-nearest neighbors (K-NNs) based on training data associated with the sub-band and on the signal to interference and noise ratio (SINR) of the sub-band. In aspects, the apparatus can identify a channel quality indicator (CQI) associated with the lowest error rate for the k-NNs and provide the identified CQI to a base station. In aspects, a neural network (NN) can provide labels for CQIs that indicate probability of choosing a CQI, and the CQI having highest probability will be provided to a base station. In aspects, a covariance matrix based on samples of a communication channel can be provided to a NN to determine a rank indicator (RI) corresponding to the channel, and channel state information associated with the (RI) can be sent to the base station. Other aspects are described.