The most valuable part of machine learning is building and effective "Predictive model". This is the development of models that get trained on prior data and then make predictions on new (unseen) data. Here is my general rule of thumb process when working with both datasets and algorithms:
1. Pick one group:
(a) Data
(b) Algorithms
(c) Parameter Tuning
2. Pick one method from the group.
3. Pick one thing to try of the chosen method.
4. Compare the results, keep if there is an improvement.
5. Repeat the steps again ...
1. Improve The Performance With Data
I - Getting more data
II - Selecting reasonable predictors
III - Preparing the data
IV - Resample and transform the Data
V - Consider performing feature selection
2. Improve The Performance With Algorithms
I - Evaluating the learning process with a reliable, e.g., k-fold cross-validation
II - Selecting the reasonable "Evaluation Metric"
III - Try several machine learning algorithms (i.e., don't select the algorithm blindly)
IV - Select the algorithm that produces best result
3. Improve Performance With Parameter Tuning
I - Optimizing one parameter using, e.g., "Cross Validation" parameter selection technique
II - Optimizing two parameters using, for instance, "Grid Search" method
III - Optimizing arbitrary number of parameters using, for example, "Multi Search" approach