For your first parameter, 50.5 Mbps, I suspect that's actually 50.5 Mpps (packets per second).
To first appreciate what these parameters mean, let's consider what "wire speed" Ethernet parameters would be. For traffic moving in one direction, on a gig link, the bandwidth would be 1 Gbps and the PPS would vary based on packet size. Often PPS is quoted for minimum sized Ethernet packets (64 bytes), which for gig would be (about) 1.488 Mpps. (NB: the PPS rate drops [much] as packet size increases [so much so, in ye olden times, switch performance was often quoted for 1500 byte packets, as often switches could not do wire speed for too much minimum size Ethernet packets], it also varies if the frame is VLAN tagged.) So, giving the foregoing, 50.5 Mpps could push about 34 Gbps minimum sized Ethernet packets.
Also with bandwidth, don't forget current Ethernet is often full duplex, meaning a gig link can carry 2 gig of traffic and would need twice the PPS rate to accomplish that. However, on swtiches, one port's ingress is another port's egress, so if you had 12 gig ports you would need about 18 Mpps. You would also need a switch fabric that supports 24 Gbps of bandwidth. (Sometimes switch fabric's internal bandwidth support cannot support all the port's bandwidth, concurrently.)
The "switching capacity is 48 Gbps" often represents the internal switch fabric capacity. Again, to support all ports concurrently, as their full capacity (which is an unusually loaded switch, at least within typical Enterprises), you need all the (full duplex) bandwidth port capacity. (Oh, sometimes fabric capacity is more than what your port's actually need. I suspect this is due to the same internal architecture being used for different switch models in the same product line. E.g. a 24 vs. 48 port model, in the same series. However, the excess capacity won't help performance any, although insufficient capacity might impede the switch's performance.)
Lastly, the "stack bandwidth" reflects "somewhat" the bandwidth of the stack ring, which can be, a bit creative, in how it's defined. Usually, ring bandwidth, like fabric bandwidth, is quoted for its full duplex capacity. (It's like saying a gig link is 2 gig, although from switch A to switch B, you'll only obtain 1 gig.) Also, how the switch puts traffic on the ring, can vary. For example, on the original 3750 series, all of a switch's fabric traffic was placed onto the ring, even if the traffic was local to just that switch member. Further, the switch that placed the traffic on the ring was also the switch that took it back off the ring (i.e. the destination switch did not remove it). With the 3750E series, unicast traffic was only placed on the ring if it had to go off the switch and the destination switch took the traffic off the ring. (Both feature were/are a huge improvement in stack ring bandwidth management. [Those switches also offered twice the raw ring bandwidth capacity too.]) I recall the 3850 supports 3 stack cables, so anyone cable might be equivalent to an 80 gig (full duplex) link.
Also somewhat unclear, is whether or how the switch's PPS and fabric's bandwidth capacity interrelate with the stack ring.
In any case, for when it was introduced, the 3850's performance was generally quite good, especially as it's designed for edge or small network usage. (I.e. heavy usage switches were usually better served by Cisco chassis switches.)
The other issue, though, with a 3850 (and often with any low end switches), those switches might be "short" compared in other high end switches with regard to internal architecture (e.g. the huge difference in StackWise vs. StackWisePlus - i..e the original 3750 vs. 3750E), feature support, and/or other hardware resources (Catalyst 3K switches often were "light" on buffer RAM compared to Catalyst 4K and/or 6K switches). I.e. even if you fully understand PPS and bandwidth specifications, there's often much more that determines whether a particular device is "best" for your intended usage.
