Which Machine Learning Algorithm Is Right for You?
| Algorithm | Dataset What is the ideal dataset size for each algorithm? |
Training Speed How quickly will the algorithm train without acceleration hardware? |
Interpretability How hard is it to see how the algorithm arrived at a decision? |
Tuning How much tuning does the algorithm allow? |
Comments |
|---|---|---|---|---|---|
| Linear models | Small | Very fast | Easy | Minimal | Widely used basic algorithm Linear SVM handles high-dimensional data well |
| Decision trees | Small | Very fast | Easy | Some | Good generalist algorithm, check for overfitting |
| (Nonlinear) Support vector machine | Medium sized | Moderately slow | Difficult | Some | Good accuracy |
| Nearest neighbor | Medium sized | Moderately fast | Moderately easy | Minimal | Lower accuracy, but easy to use and interpret |
| Naïve Bayes | Medium sized | Very fast | Moderately easy | Some | Widely used for text analytics (e.g., spam filtering); kernel Bayes will run slower |
| Ensembles | Large | Moderately fast | Difficult | Some | Higher accuracy with a tradeoff of lower interpretability |
| Neural network (shallow) | Medium sized | Moderately fast | Moderately easy | Some | Still used for signal classification, compression, and forecasting |
| Deep nets | Large | Very slow | Difficult | A lot | A standard algorithm for image, video, signals, and text |
