If the purpose of the analysis is to create a model for prediction or retrodiction and if your data is discrete integers and random, then there are statistical tools available to determine if it is close to any of the many statistical distributions. If so, then the data may be predictive of the population. Since the data is not time series, the predictions might be related to some other parameter. If the data is chaotic, you may be able to find some quasi-linear patterns within the chaotic behavior that are locally predictive around or near your data set (but not globally predictive). Alternatively, if the deterministic and random characteristics of the data are muddled, you may try mapping "phase" spaces to see if there are distinguishing elements. ("Phase" here does not necessarily imply temporal data).

If the data are random integers, the statistical modeling of the data should make sense. For example heart rate variability data consists of R-R intervals which are random integers (being the time difference in milli seconds between two consecutive heart beats), I have earlier applied stochastic fuzzy modeling techniques for the analysis of R-R intervals. If you prefer a deterministic approach, then e.g. deep autoencoder based analysis could be carried out. Hope it helps!

The first step is using plots and the nature of variables. What do the plots imply? The variation and the pattern of the well-known indices like mean, variance, covariance (auto-covariance), autocorrelation, and asymptotic relations are beneficial as well. Finally, you can model discrete outputs through continuous random processes either using functions like round and floor or without transforming the outcomes.