Human immunodeficiency virus (HIV-1) is responsible for acquired immune deficiency syndrome in humans. Despite the advances of antiretroviral therapy, persistence of the virus due to establishment of transcriptional latency still remains a barrier for the cure. HIV-1 integration presents as a complex interplay between the virus, cell chromatin and nuclear organisation and is influenced by a myriad of factors. HIV-1 preferentially integrates actively transcribed genes rich in active histone marks and spatially connected genomic compartments with more super-enhancers. Machine learning methods are a good choice for analyses of such problems as they allow for an exploration of a multitude of factors and they can discern complex patterns in big biological datasets, as well as provide insight into relative importance of factors. Model developed in this paper achieves good discrimination of recurrently integrated genes, confirms previous findings and quantifies the importance of spatial data in predicting HIV-1 integrations compared to gene expression, histone modifications and chromatin state. Variable importance analysis, using a permutation test, determines higher gene expression levels and active histone marks (H3K27ac and H3K36me3 in gene bodies, and H3K4me3 in transcription start site neighbourhood) as main factors in discrimination of recurrently integrated genes from genes without integrations. ROC curve analysis reveals that the spatial division of genome into compartments and localisation of super-enhancers inside them significantly contribute to discrimination (p < 0.05), independently of histone marks and gene expression. It is also found that removal of gene expression data from the model does not lead to information loss that would be expected from relative importances determined by the permutation test. Computational methods and predictive modeling lead to consistent conclusions and provide useful tools for retroviral integration study. Similar methods could be used in order to infer differences in mechanisms of integration site selection in different infection phases and clarify the formation of a latent HIV-1 reservoir, which could subsequently be used to advance antiretroviral therapy development. Also, herein developed and described computational analysis methods can be useful for further research of many aspects of retroviral DNA integrations, as well as other biological problems.