How to Detect Data Drift in Machine Learning Models
Production monitoring, statistical tests, drift signals, and the distinction between detectable distribution movement and model failure.
Statistics, Machine Learning & Applied Mathematics
Research, technical articles and open-source software on statistical modelling, machine learning, time series, data systems and applied mathematics.
I write for practitioners who need more than a generic tutorial: assumptions, diagnostics, equations, code, failure modes, and the judgment required to use methods responsibly.
About and editorial standards Explore the archive Papers and research Projects and packages
Selected work
These are better entry points than the chronological archive.
Probability Calibration in Machine Learning
Why probability outputs need calibration, how to score them, and where Venn-Abers style predictors fit.
Forecasting Baselines That Are Hard to Beat
A practical argument for naive and seasonal baselines before treating any forecasting metric as evidence of skill.
Conformal Prediction for Operational Risk Decisions
Prediction intervals connected to decisions, not just uncertainty decoration.
Explore by topic
Statistics & Probability
Inference, modelling, probability, diagnostics, survival analysis, robust methods and uncertainty.
Machine Learning
Model evaluation, monitoring, calibration, drift, tabular learning, data quality and MLOps.
Time Series & Forecasting
Forecasting, baselines, seasonality, anomaly detection, feature engineering and state-space models.
Mathematics
Applied mathematics, stochastic processes, optimization, graph theory and the foundations behind data science.
Data Engineering & Systems
Pipelines, monitoring, real-time processing, software practice and production data systems.
Research Methods & Causal Inference
Experimental design, causal reasoning, preregistration, measurement, ethics and research policy.
Open-source projects
The software pages are technical assets, not just documentation links. The Python portfolio is published on PyPI and spans scientific machine learning, survival simulation, QCA, time-series representations and WiFi activity recognition.
- PyPI package portfolio — six Python packages with install commands, documentation and source links.
- genSurvPy / gen-surv — survival-data simulation and visualization for statistical research and benchmarking.
- unconfoundedr — R tools for comparing randomized and observational estimands under confounding and transportability concerns.
Recent writing
Information Geometry for Data Science: Curvature, Models, and Learning
Information geometry treats probability models as geometric objects, making it easier to reason about distance, curvature, uncertainty, and learning.
Discrete Mathematics for Data Science: States, Constraints, and Algorithms
Discrete mathematics is the part of mathematics that explains how data systems make decisions, count possibilities, represent relationships, and enforce constraints.
Bayesian Decision Theory for Data Science: From Uncertainty to Action
Bayesian decision theory connects statistical uncertainty to action by asking not only what is likely, but what decision is best under uncertainty.
Evaluating the ROI of Predictive Maintenance: A Practical Measurement Framework
Predictive maintenance only creates value when better predictions change maintenance decisions. This article explains how to measure that value without confusing model performance with business impact.
Data Visualization and Dashboards for Predictive Maintenance
Predictive maintenance dashboards should not merely display sensor data. They should help teams decide what to inspect, when to act, and which risks matter most.
Cloud Computing and Edge Analytics in Predictive Maintenance
Predictive maintenance systems rarely live entirely in the cloud or entirely at the edge. Effective architectures split work across sensors, gateways, plant systems, and cloud platforms.
State Space Models and the Kalman Filter
The Kalman filter is usually introduced as a tracking algorithm for spacecraft. It is more useful understood as the general engine for estimating hidden state from noisy observation.
Measurement Invariance for Machine Learning Monitoring
Learn how measurement invariance gives model monitoring teams a statistical language for detecting when features, labels, or scores stop meaning the same thing across time or groups.
Global vs Local Models in Time Series Forecasting
The traditional approach fits one model per series. Modern practice often fits a single model across thousands of them, and usually wins.
Causal Feature Selection for Observational Machine Learning
Predictive feature selection is not enough when a model supports interventions. This article explains how causal thinking improves feature design in observational machine learning.