Technically (and sonically with high fidelity samples and good monitoring) upsampling from 44.1K to 48K sounds worse than downsampling from 48K to 44.1K.
Why? When you downsample, you are throwing some data away, since a 48K sample has more data than a 44.1K sample. But upsampling from 44.1K to 48K, you are having to interpolate and actually create data to make up the difference.
It really comes down to if you like the way it sounds, or not, since grainier or grittier may be just the thing. Or maybe not.
Using a high quality SR like R8Brain to batch convert, as suggested above, will result in better quality conversion. Only you can decide if it’s worth hassle to you. R8Brain is free. Take a few samples, convert both ways and compare. Great learning process.