Take a look at this thread. Some forum members were creating CoinJoin transactions with each other. Review how the Wasabi Wallet works; it combines inputs from various users to obfuscate the relationship between each user's inputs and outputs. At one point, now-defunct exchange Mt Gox allowed users to 'import' their private keys into their accounts, and Mt Gox would use those private keys to create transactions that spent any unspent outputs spendable with the private keys, and the transactions would be spending unspent outputs of other customer's private keys, and some of Mt Gox's own coin. With the advent of Lightning Network, many people will sign transactions that have other people's inputs.

The point of all this is that a single transaction that connects two addresses together is not necessarily enough to link two businesses together, absent additional evidence.

Also, a classification model that is accurate 96% of the time (it is unclear how you are measuring accuracy) has very high accuracy. My first reaction to that high of claimed accuracy is that you might have data leakage. I can't point to the source without looking at your specific steps to train your model, which understandably may not be something you want to share.