Putting PSO inside the experiment loop
Why optimization is most useful when it improves the experiment process, not only the final accuracy number.
Hyperparameters are coupled decisions
Learning rate, batch size, dropout, and hidden-unit choices do not act independently. A value that works in one configuration can behave differently when the rest of the training setup changes.
Particle Swarm Optimization provides a structured way to search those combinations. Each candidate carries information from its own best result and from the strongest result found by the group.
The objective must reflect the problem
For multi-class disease classification, accuracy alone can hide weak performance on individual classes. The experiment loop should retain precision, recall, F1-score, and class-level diagnostics alongside the optimization target.
That makes the search auditable. A better candidate is valuable because its behavior is understood, not simply because one number increased.
Optimization is an engineering system
The reusable part of the research is the loop around the model: define a search space, run comparable trials, preserve results, and inspect the tradeoffs. PSO is one strategy inside that system. Reproducible experiments are what make the result useful beyond a single paper.