If a channel has a much better balance of incoming and outgoing flow which can be tweaked by changing the rates so that it never gets too heavy on one or the other side, then the risks go way down, and the necessary fee rates tends towards zero, because eventually you'll move enough traffic to cover all the costs over time. These types of channels are the healthiest and most sustainable.

And it's practically impossible to predict or otherwise plan ahead to choose channels that will likely have such bi-directional flow. Experimenting on LN is expensive, because channel closes are expensive. And people generally don't have or want to risk tons of BTC for this purpose, even if it's potentially profitable. I wish I knew how many LN nodes out there were profitable, but I would guess close to 0.01%. Because it's a very hard problem to solve which channels are the most ideal to open and will have sufficient bi-directional flow. And you can't really solve that by simply opening up channels in one particular way, because I believe, over time, it will always tend towards an unbalanced situation, at which point, the only real rebalancing option you have is to close the channels that are stuck at 99% and 1% and haven't moved any sats.

• Take the network graph from my node, and apply some filters
• Analyze, for a given payment size, how many nodes it can reach in both directions (inbound, outbound) for a given max cost.
• Then, for each candidate node, look how many more nodes can be reached (cost becomes less than the max cost, or nodes that could be reached but cost is now slower) in each direction
• For each candidate, i take their 'new node reach' in the worst direction, order that
• And finally select the one that will bring more nodes 'in reach' in the 'worst direction'.