Description
Finds the linear combinations of behavioural metrics that best separate known user groups, classifying new users and showing which variables discriminate most.
Linear discriminant analysis (LDA) finds the linear combinations of predictor variables that maximise the ratio of between-group to within-group variance for a set of known groups. Applied to UX, the predictors are behavioural metrics (feature use, session frequency, error counts, time on task) and the groups are meaningful labels such as retained versus churned or novice versus expert. LDA both classifies new users and, through its standardised coefficients, explains which behaviours most strongly separate groups, which makes it more interpretable than black-box classifiers. Rigour comes from checking assumptions (approximate multivariate normality, homogeneity of covariance via Box's M), using cross-validated rather than resubstitution accuracy, and reporting chance-corrected performance against base rates.
Attributes
- Discipline
- Statistics & Modelling
- Loss type
- Siloed
- Stage
- Definition, Post-launch
- Data
- Quantitative
- Effort
- Days
Steps
- Label a subset of sessions with behavioral categories (ground truth from qualitative analysis or screen recording review).
- Extract behavioral features for all sessions.
- Fit LDA: find the discriminant function that maximally separates labeled classes.
- Validate: leave-one-out cross-validation accuracy.
- Extract discriminant loadings: which features most separate the classes?
- High-loading features = the behavioral markers most diagnostic of the classification.
Sources
- Pattern Classification — Richard O. Duda
- lifelines (survival)
- scikit-learn (GMM, LDA, UMAP)
- pandas (log parsing)
- matplotlib
- plotly
- Fisher, R. A. (1936) The use of multiple measurements in taxonomic problems. Annals of Eugenics
- Hastie, T., Tibshirani, R. & Friedman, J. (2009) The Elements of Statistical Learning, 2nd ed.
- The Inquiry Series, C4 — Behavioural Log Analysis — From Clickstreams to Personas
- Pattern Classification (Richard Duda, Peter Hart & David Stork)
- Practical Statistics for Data Scientists (Peter Bruce, Andrew Bruce & Peter Gedeck)
- Methods Atlas (Vivek Gopinath)
- Fisher (1936) The use of multiple measurements in taxonomic problems
- Pattern Classification (Duda)
- Practical Statistics for Data Scientists
- Methods Atlas (Gopinath)
- The Inquiry Series, Ch 54 — Linear Discriminant Analysis & Behavioral Classification
- Factor and Discriminant Analysis (Sauro & Lewis, MeasuringU training)
- Practical Statistics for Data Scientists (Bruce & Bruce)
Also known as
- LDA
- Fisher's discriminant
- Discriminant function analysis
- Fisher's linear discriminant
- Linear discriminant analysis
- Fisher's discriminant analysis
- LDA for supervised classification of session data
- Linear Discriminant Analysis for Behavioural Classification
Categories
- Analytics & Behaviour
- Statistics & Modelling