Genetic Algorithm

How genetic algorithms work

A genetic algorithm treats optimisation as an evolutionary search process. It starts with a population of candidate solutions, evaluates them against a fitness function, and then uses selection, crossover, and mutation to generate the next generation. The goal is not to prove the perfect answer, but to improve solution quality over repeated iterations.

This makes the method especially useful when the search space is too large, nonlinear, or poorly behaved for exhaustive search. Because it explores multiple candidates at once, it can escape some local optima that trap greedy methods, although it still depends heavily on how the problem is encoded and scored.

Core mechanics and solution encoding

The practical value of a genetic algorithm depends on how candidate solutions are represented. A chromosome can be a binary string, a vector of parameters, a route, a schedule, or another structure that can be recombined and mutated without breaking the problem constraints. If the encoding is poor, the search may spend effort on invalid or low-value candidates instead of useful ones.

Fitness evaluation is the engine of the process. It turns a business objective, cost function, error measure, or constraint score into a ranking signal that the algorithm can optimise. In many real problems, the fitness function matters more than the choice of mutation rate or crossover style, because it determines what “better” actually means.

Where genetic algorithms fit in optimisation

Genetic algorithms are a good fit for search problems with large combinatorial spaces, competing objectives, or rugged landscapes where classical optimisation struggles. They are often used as approximate methods when the acceptable answer is “good enough” rather than mathematically optimal. That makes them common in scheduling, routing, design search, feature selection, and parameter tuning.

They are usually not the best choice when the problem has a clean closed-form solution, strong convexity, or a faster specialised solver. In those cases, the overhead of population management and repeated evaluation can be unnecessary. Their strength is flexibility, not guaranteed optimality.

Limitations, tuning, and practical trade-offs

Genetic algorithms are sensitive to parameter choices such as population size, mutation rate, crossover rate, stopping criteria, and elitism. Too little variation can cause premature convergence, while too much variation can make the search chaotic and slow. The same algorithm can perform very differently across problems, so tuning and validation are part of the method, not an optional afterthought.

They also need careful handling of constraints. If invalid candidates are common, the algorithm may waste iterations exploring unusable regions unless the encoding, fitness function, or repair logic is designed well. In practice, the method is best understood as a controllable heuristic: powerful when the search space is difficult, but only as reliable as the objective and representation you give it.

Related resources from NHI Mgmt Group

• How should security teams choose a password hashing algorithm for modern applications?
• How should security teams prevent JWT algorithm confusion in verification code?
• Why do JWT algorithm confusion attacks bypass normal authentication controls?
• When does crypto-agility matter more than selecting a specific PQC algorithm?