Hyperparameter tuning is a crucial step in machine learning model development. It involves adjusting the settings of the algorithm, which are not learned from the data, to improve model's performance. Here are the main techniques used:

1. Grid Search: This method involves specifying a set of possible values for each hyperparameter. The algorithm then systematically tries all combinations of values to find the best ones.
2. Random Search: Instead of trying all combinations, like in grid search, this method randomly selects values for each hyperparameter and evaluates the model's performance. This approach can be more efficient when dealing with a large number of hyperparameters.
3. Bayesian Optimization: This method tries to find the optimum by constructing a posterior distribution of functions (Gaussian process) that best describes the function you want to optimize. As points are sampled, the method updates the posterior.
4. Gradient-based Optimization: This method uses gradient information from the previous steps to guide the search for optimum hyperparameters.
5. Evolutionary Algorithms: These are based on the principles of evolution and genetics. The method starts with a population of random solutions and evolves them by using operations such as mutation, crossover (recombination), and selection.

Remember, hyperparameter tuning is part art, part science and the results can significantly impact the predictive power of your model.