EXIN BCS ML Award study plan

Define supervised, unsupervised, and reinforcement learning and classify ten given problem statements into the correct paradigm with a one-line justification each.

Distinguish classification from regression and identify which evaluation metric family applies to each, writing the rule in your own notes.

Set up a working Python environment with NumPy, pandas, scikit-learn, and matplotlib, and confirm imports succeed in a single script.

Write and run Python covering variables, lists, dictionaries, loops, functions, and list comprehensions, producing at least ten short exercises you authored yourself.

Explain the difference between training, validation, and test data and why leakage invalidates results.

Load a CSV with pandas, inspect shape, dtypes, and missing values, and document five concrete cleaning decisions you made.

Compute and interpret mean, median, standard deviation, and correlation for numeric columns, and state which features look informative.

Implement a train/test split manually and with scikit-learn, and explain why stratification matters for imbalanced classes.

Describe the bias-variance trade-off and diagnose underfitting versus overfitting from train and test scores you produce.

Summarise the main stages of a machine-learning project lifecycle from problem framing to monitoring.

Implement linear regression with scikit-learn, report coefficients and R-squared, and interpret each coefficient in plain language.

Implement logistic regression for binary classification and explain why accuracy alone is misleading on imbalanced data.

Derive and apply precision, recall, F1, and the confusion matrix to your model's predictions.

Plot and interpret an ROC curve and compute AUC, stating what the value means for a non-technical stakeholder.

Explain how gradient descent updates parameters and the role of the learning rate.

Compare k-nearest neighbours and decision trees on the same dataset and explain how each handles feature scaling.

Describe how a decision tree chooses splits using impurity measures, and explain pruning to control overfitting.

Explain ensemble methods including random forests and gradient boosting at a conceptual level, and list two trade-offs versus a single tree.

Implement k-fold cross-validation and report mean and standard deviation of scores.

Apply at least two hyperparameters searches and justify the chosen settings.

Explain k-means clustering, its assumptions, and how the number of clusters can be chosen using an elbow plot or silhouette score.

Describe principal component analysis as dimensionality reduction and interpret explained variance ratios you compute.

Contrast supervised and unsupervised evaluation and explain why clustering lacks a single ground-truth metric.

Explain the purpose of support vector machines and the effect of the kernel choice at a conceptual level.

Identify which algorithm families suit tabular, text, and image data respectively, with a stated reason for each.

Build a preprocessing pipeline with scikit-learn that handles missing values, scaling, and categorical encoding without leakage.

Engineer at least three new features and measure whether each improves cross-validated performance.

Explain feature importance from a tree-based model and state two limitations of that interpretation.

Document ethical risks in a chosen use case, covering bias, privacy, and transparency concerns.

Describe the requirements of data protection rules such as GDPR at a level relevant to handling personal data in a model.

Explain how a trained model is deployed and monitored, including data drift detection and retraining triggers.

Distinguish model performance metrics from business metrics and map one model to three business KPIs.

Describe the roles of data scientist, ML engineer, and domain expert across a project lifecycle.

Analyse two case studies of ML failure or harm and extract one preventive practice from each.

Draft a model card summarising intended use, limitations, and evaluation results for your week 6 model.

Answer at least 120 self-written or freely available practice questions across all four domains and log accuracy per domain.

Re-derive the weakest two domain summaries from memory and correct them against your notes.

Explain aloud, without notes, ten core concepts spanning paradigms, algorithms, evaluation, and deployment.

Review every term in your week 1 glossary and rewrite any definition you can no longer justify.

Complete one timed mock covering all domains and review every wrong answer to its root cause.