AAI_2025_Capstone_Chronicles_Combined

Breast Tumor Classification Using Quantum Neural Networks

20

Figure 14

Final and best result: feature selection, diƯerent quantum feature mapping, and 8 hidden layers improving performance to ~90%.

Areas for Future Work

As briefly mentioned previously, an area for further exploration would be attempting a

wider range of changes to different model components and exhaustively testing them. As a brief

example, different combinations of batch size and epochs could be attempted. With batch sizes

of 8, 16, 32 and 5, 10, or 20 epochs there would be 9 combinations: batch of 8 and epoch of 5,

batch of 8 and epoch of 10, batch of 8 and epoch of 20, batch of 16 and epoch of 5, …, batch of

32 and epoch of 20. The number of combinations quickly increases as other components are

added and wider ranges are attempted- making this impractical from a computation perspective

at some point. Instead of this exhaustive brute force approach, another alternative would be to

apply genetic programing techniques. That entails basically selecting random combinations

(genomes) in the search space, seeing which perform best (fitness function), and then

combining the best ones with random changes (mating and mutation), to see if the next

generation performs better.

273