Machine Learning
MediSight AI
An explainable healthcare risk-modelling workflow built around model evaluation, experiment tracking, and a clinician-facing interface.
Case study · Machine Learning
Customer Churn Prediction
Classification · Coursework
A supervised-learning study comparing classification approaches for identifying customers who may leave a service.
01 / Context
The problem
Retention teams need useful risk signals, but classifying churn requires more than selecting a model with a high headline accuracy. Precision, recall, F1, class balance, and the decision threshold all matter.
My role: Team coursework project; model experimentation and analysis are represented in the public repository.
02 / Implementation
How it works
01Understand the data
Inspect feature types, missingness, distributions, and the churn label before fitting models.
02Build baselines
Compare a linear model with tree-based and neural approaches to understand trade-offs in performance and interpretability.
03Evaluate the right things
Report accuracy alongside F1 and other class-sensitive metrics so the model is not judged on a single number.
04Keep the output actionable
The project is an educational prediction workflow; it does not claim to be a production retention platform.
Implementation highlights
- Data cleaning and exploratory analysis
- Compared Logistic Regression, Random Forest, and an MLP classifier
- Explored clustering as a complementary unsupervised view of customer groups
- Selected Logistic Regression for the reported baseline result
03 / Evidence
What can be verified
The points below summarize the public repository and owner-provided CV where noted. Read the source for implementation details and the evaluation setup before drawing further conclusions.
Accuracy84%
Owner-confirmed CV figure; verify exact split and evaluation setup in the repository before direct comparisons.
F1 score71%
Owner-confirmed CV figure; F1 depends on the positive class and decision threshold.
Selected baselineLogistic Regression
Reported as the selected model in the CV.
04 / Boundaries
Limitations and next questions
- The reported values need to be interpreted in the context of the data split and class distribution.
- A coursework result is not evidence of production performance or causal drivers of churn.
- Clustering is exploratory and should not be interpreted as ground-truth customer segments without validation.
Evidence-first note: a live demo demonstrates an interface and workflow; it does not by itself validate model quality, security, or real-world impact.
Continue exploring