Advanced Data Analysis Using R Training Course

Participants can import, standardise and combine data from common analytical source formats.

• Readr import specifications for CSV, Excel and delimited files
• JSON parsing from web APIs using httr2 and jsonlite
• Database connections through DBI and parameterised SQL queries
• Character encoding, date parsing and locale settings during import
• Schema comparison across monthly extracts and historical source files
• Source metadata recorded for ownership, refresh frequency and limitations
• API pagination, rate limits and error handling patterns

Participants can reshape and transform complex datasets into analysis-ready tables using tidyverse methods.

• Tidy data principles for observations, variables and analytical units
• dplyr pipelines for filtering, recoding, grouping and summarising records
• pivot_longer and pivot_wider for reshaping repeated measurement data
• Conditional logic using case_when, if_else and coalesce
• Window functions for ranks, lags, rolling values and cohorts
• Date intervals calculated with lubridate for operational measures
• Functions reduce repeated transformation logic across related datasets

Participants can join, query and validate relational data without losing analytical control.

• Primary keys, foreign keys and grain defined before joining tables
• Inner, left, right and full joins compared for business questions
• Join cardinality checks identify unintended row multiplication risks
• SQL common table expressions organise multi-stage database queries
• Window functions calculate customer, account and transaction histories
• dbplyr translates dplyr pipelines into database-native SQL
• Query plans reviewed to identify inefficient database operations

Participants can assess data quality, document exceptions and select appropriate missing-data treatments.

• Validation rules test ranges, formats, uniqueness and referential integrity
• Missingness profiles distinguish MCAR, MAR and MNAR assumptions
• Outlier detection using IQR rules, z-scores and robust statistics
• Duplicate record investigations separate repeated events from data errors
• Audit tables capture rejected records, reasons and corrective actions
• Multiple imputation with mice compared against complete-case analysis
• Quality checks embedded within reproducible transformation pipelines

Participants can identify patterns, anomalies and plausible relationships before formal model development.

• Distribution summaries using quantiles, skewness and robust spread measures
• Grouped profiling reveals variation by segment, period and location
• Correlation matrices screened for multicollinearity and spurious associations
• Faceted plots compare distributions across categories and cohorts
• Bivariate analysis separates association, confounding and causal claims
• Sampling strategies inspect large datasets without overwhelming visual outputs
• Exploration logs record hypotheses, anomalies and follow-up questions

Participants can produce clear statistical graphics that support analytical interpretation and decisions.

• Grammar of Graphics layers map data, aesthetics and geometries
• Scales and coordinate systems prevent misleading visual comparisons
• Confidence intervals displayed with point estimates and uncertainty bands
• Annotation techniques identify targets, thresholds and material exceptions
• Colour palettes selected for accessibility and categorical discrimination
• Small multiples compare trends across regions, products and cohorts
• Publication-ready themes standardise fonts, labels and chart elements

Participants can select and interpret inferential methods for comparisons, associations and controlled tests.

• Sampling distributions link observed statistics to population uncertainty
• Confidence intervals interpreted beyond simple statistical significance claims
• t-tests, ANOVA and non-parametric alternatives selected by assumptions
• Chi-square tests assess association between categorical variables
• Power analysis estimates sample sizes for planned comparisons
• A/B test design addresses randomisation, exposure and outcome definitions
• Multiple-testing adjustments control false discoveries across repeated comparisons

Participants can build, diagnose and interpret multiple linear regression models in R.

• Model specifications connect business hypotheses with measurable variables
• Multiple regression estimates adjusted relationships among predictor variables
• Interaction terms test whether effects differ across defined groups
• Residual plots diagnose linearity, normality and constant variance
• Variance inflation factors identify problematic multicollinearity
• Influence measures examine leverage, Cook's distance and influential cases
• Coefficient tables translated into operational and financial implications

Participants can model binary, count and hierarchical outcomes using suitable generalised model forms.

• Logistic regression estimates probability for binary outcome events
• Odds ratios, marginal effects and predicted probabilities interpreted correctly
• Poisson and negative binomial models handle count outcomes
• Offset terms model rates using differing exposure periods
• Mixed-effects models account for clustered branches, sites and teams
• Random intercepts and slopes represent group-level variation
• Overdispersion and zero inflation checked before interpreting count models

Participants can prepare model features and estimate generalisation performance using resampling methods.

• Train, validation and test partitions prevent optimistic performance estimates
• Cross-validation evaluates models across repeated training data splits
• Stratified sampling preserves rare outcome proportions during partitioning
• Recipes standardise transformations within tidymodels modelling workflows
• Dummy encoding, normalisation and transformations fitted without data leakage
• Class imbalance addressed through weights, downsampling and SMOTE
• Feature creation uses domain logic, lags, ratios and aggregation

Participants can train, tune and compare tree-based models for classification and regression tasks.

• Decision trees partition data using interpretable split rules
• Random forests reduce variance through bootstrap aggregation and random features
• Gradient boosting builds sequential trees that reduce residual error
• Hyperparameter tuning selects depth, learning rate and tree counts
• ROC curves, precision-recall curves and threshold selection compared
• Regression metrics include RMSE, MAE and explained variance
• Model comparison tables balance accuracy, stability and interpretability

Participants can develop and assess segments using clustering and dimension-reduction techniques.

• Standardisation prevents high-scale variables dominating distance calculations
• K-means clustering groups records around centroid-based similarity patterns
• Hierarchical clustering reveals nested relationships among observations
• Silhouette scores assess cluster separation and assignment quality
• Principal component analysis reduces correlated variables into components
• Cluster profiles translate statistical groups into operational segments
• Segmentation stability tested across samples and model specifications

Participants can analyse time-dependent data and produce validated forecasts for planning decisions.

• Time indexes checked for gaps, duplicates and irregular frequency
• Trend, seasonality and calendar effects separated using decomposition
• Autocorrelation plots identify dependence across prior time periods
• ARIMA models selected through differencing, diagnostics and information criteria
• Exponential smoothing forecasts level, trend and seasonal patterns
• Rolling-origin resampling evaluates forecast performance over time
• Prediction intervals communicate uncertainty in demand and budget forecasts

Participants can prepare, quantify and model text data from surveys, cases and documents.

• Text cleaning handles case, punctuation, stopwords and tokenisation
• Regular expressions extract identifiers, dates and structured text patterns
• Term frequency-inverse document frequency identifies distinctive language
• Sentiment lexicons score positive, negative and emotion-related terms
• Topic models identify recurring themes within large document collections
• N-grams reveal common phrases and sequential language patterns
• Manual coding samples validate automated text classification outputs

Participants can analyse and visualise location-based patterns using vector spatial data in R.

• Simple Features objects store points, lines, polygons and attributes
• Coordinate reference systems selected before distance and area calculations
• Spatial joins assign events to branches, districts and service areas
• Choropleth maps compare normalised rates rather than raw counts
• Buffer analysis identifies records within defined travel-distance thresholds
• Spatial aggregation summarises transactions, incidents and demand by area
• Map classifications documented to prevent misleading geographic comparisons

Participants can explain model behaviour, assess limitations and document material analytical risks.

• Partial dependence plots show average predictor-response relationships
• Permutation importance ranks variables by contribution to model performance
• SHAP values explain individual predictions and local feature effects
• Calibration curves compare predicted probabilities with observed outcomes
• Fairness metrics compare error rates across protected or relevant groups
• Sensitivity analysis tests conclusions against assumptions and parameter choices
• Model cards document intended use, limitations, monitoring and approvals

Participants can publish reproducible analytical reports and prototype interactive decision-support outputs.

• Quarto documents combine narrative, code, tables and visual evidence
• Parameterized reports generate consistent outputs for multiple business units
• Inline R code keeps reported figures aligned with calculations
• gt tables format model results, exceptions and management metrics
• Shiny reactive inputs support controlled exploration of analytical scenarios
• Report structures distinguish findings, limitations, recommendations and appendices
• Publication checklists verify reproducibility, confidentiality and presentation quality

Participants can build, review and present a complete R analysis project from source data to decision recommendation.

• Frame a decision question, analytical scope and measurable success criteria
• Prepare source data with documented transformations and quality controls
• Select appropriate exploratory, inferential or predictive modelling methods
• Validate findings using diagnostics, resampling and sensitivity checks
• Produce a Quarto report with charts, results and recommendations
• Review peer projects against reproducibility, validity and communication criteria
• Present methods, findings, limitations and proposed next actions