Writing operating-system kernels.

Creating blockchain wallets only.

Replacing all Python libraries in production systems.

Visual data analysis and workflow-based machine learning experiments.

Changing the target variable after seeing the test score.

Publishing a model without validation.

Checking dataset size, feature types, missing values, and distributions.

Removing all plots and statistics from the report.

Predicting house prices from historical sales data.

Classifying emails as spam or not spam using labeled examples.

Grouping customers into segments based on behavior patterns.

Estimating tomorrow's temperature using past weather labels.

Mean absolute error

R-squared only

F1-score

Davies-Bouldin index only

Use accuracy only and ignore the minority class.

Train only on the test set.

Delete all fraud cases because they are rare.

Use class weights, resampling, SMOTE, and metrics such as precision, recall, F1, or ROC-AUC.

To train a neural network without a loss function.

To increase the number of target classes.

To remove all categorical features from a dataset.

To transform features to have approximately zero mean and unit variance.

Replace them with column names.

Convert all missing values into target labels.

Impute them using mean, median, or a model-based method.

Ignore them in all algorithms without checking.

Only association-rule transactions.

Sequential data such as text, signals, or time series.

Only geospatial polygons without attributes.

Only static unordered columns with no sequence.

Using only the first row of the dataset.

Creating meaningful variables from raw data, such as extracting hour-of-day from timestamps.

Copying the target variable into the input features before testing.

Deleting all columns without explanation.

A database normalization rule.

A reinforcement-learning reward function.

A type of convolutional image filter.

A weighting scheme that reflects word importance in a document relative to a corpus.