Data quality for machine learning
Machine learning results are only as useful as the table behind them. Before training a CSV model, review missing values, schema consistency, target leakage risk, identifier-like columns, high-cardinality fields, target imbalance, and sample-size limits. MLdeck Data Quality helps make those checks visible in a browser-local workflow before users move into AutoML.
Why data quality comes before model choice
A stronger algorithm cannot fix a target that leaks, a table full of identifiers, or labels that do not represent the future prediction question. Data quality for machine learning starts with the shape and meaning of the CSV: what each row represents, which columns are available at prediction time, how missing values appear, and whether the target is usable.
MLdeck treats Data Quality review as a separate step from model training. The review helps users decide whether a CSV is ready for reporting, cleanup, sharing, or browser-local AutoML. It does not replace model validation or domain review.
Signals to review before AutoML
Missing values
Missingness can reflect import issues, incomplete collection, or meaningful states. Review which columns are incomplete and whether missing values cluster in important groups.
Schema and type consistency
Mixed numeric/text values, inconsistent date formats, and confusing categories can distort preprocessing and model interpretation.
Target leakage risk
Columns that reveal the answer or come from after the prediction moment can make exploratory metrics look stronger than future behavior.
Identifier-like columns
Customer IDs, row keys, emails, order numbers, and transaction identifiers often need exclusion or careful justification before training.
High cardinality
Fields with many unique values may behave like identifiers, especially when row counts are limited or categories repeat unevenly.
Target imbalance and sample limits
Rare classes, narrow target ranges, and tiny samples can make a leaderboard hard to interpret without stronger validation evidence.
Data Quality is not model validation
Data Quality review asks whether the CSV is suitable enough to use. Model validation asks how a trained model behaves. A dataset can pass basic quality review and still produce weak models. A model can show strong exploratory metrics and still fail under stricter holdout, temporal, group-aware, or external validation.
Use Data Quality review before training, then pair AutoML results with AutoML validation evidence, baseline comparison, leakage review, and user-side verification.
How MLdeck fits this workflow
MLdeck Data Quality reviews CSV readiness signals in the browser. MLdeck AutoML can then train and compare tabular models in browser-local workflows when training is the right next step. During normal browser-local workflows, raw CSV training rows are not uploaded to MLdeck servers.
Users remain responsible for local device security, browser extensions, organizational policy, downloaded artifacts, and external systems used after export.
Data quality FAQ
What should I check before training a model?
Review missing values, schema consistency, identifier-like columns, high-cardinality fields, target leakage risk, target imbalance, and whether the sample is large and representative enough for the question.
Is data quality the same as model validation?
No. Data Quality review checks whether the CSV looks suitable for downstream work. Model validation evaluates trained model behavior with metrics, holdouts, baselines, and external review.
Does the CSV data quality checker certify a dataset?
No. It provides browser-local readiness signals and recommendations. It does not certify correctness, compliance, privacy status, or model quality.
Should I use Data Quality before AutoML?
Yes, when the goal is to understand whether a CSV is ready for reporting, cleanup, sharing, or model training.
Continue the review
Use these pages to connect data quality review with browser-local model training and examples.