Despite rapid development of laser powder bed fusion (L-PBF) and its monitoring techniques, there is still a lack of in situ crack detection methods, among which acoustic emission (AE) is one of the most sensitive. To elaborate on this topic, in situ AE monitoring was applied to L-PBF manufacturing of a high-strength Al-based alloy and combined with subsequent X-ray computed tomography. By using a structure borne high-frequency sensor it was possible to detect AE activity associated with cracking, which occurred not only during L-PBF itself, but also after the build job was completed, i.e. in the cooling phase. AE data analysis revealed that crack-related signals can easily be separated from the background noise (e.g. inert gas circulation pump) through their specific shape of a waveform, as well as their energy, skewness and kurtosis. Thus, AE was verified to be a promising method for L-PBF monitoring, enabling to detect formation of cracks regardless of their spatial and temporal occurrence. Further, Machine learning (ML) was implemented in order to automate differentiation of crack AE events from background noise. For this purpose, AE events were investigated as finite-dimension vectors in the spaces of standardized moments and principal components (PC) of waveforms and spectra. Introduced vectors were used as input for ML binary classification models (Logistic Regression, Support Vector Machine, Random Forest, and Gaussian Process Classifier). The ML models reached the highest classification accuracy, up to 99 %, for events represented in the space of spectra PC. The findings of this work will be used as a basis for the design of more sophisticated systems of AE events differentiation in multiclass classification setup.