Updated November 3, 2026
A machine-learning model can only learn from the information and patterns represented in its training data. Before selecting algorithms, understand what the dataset contains, how it was collected and which quality issues may affect the result.
Start by understanding the dataset
Inspect row and column counts, data types, missing values, duplicate records, target distribution and suspicious values. These checks help reveal whether the dataset matches the problem you want to solve.
Do not assume that every missing value is an error or that every duplicate should be removed. Decisions depend on how the data was collected and what each row represents.
Treat cleaning as a reviewed transformation
Preview transformations before applying them. Keep the source unchanged and create a separate cleaned output so the effect of each operation can be inspected and reproduced.
Protect evaluation from leakage
Fit preprocessing steps only on training data within the validation pipeline. Information from validation or test data must not influence training-time transformations or feature selection.
Choose a split strategy that reflects the real task. Time-dependent data generally needs time-aware validation rather than a random split.
Use readiness scores carefully
A readiness assessment can summarize data-quality risks and point to next steps. It is a heuristic, not a guarantee of accuracy, fairness or production performance.