4D Training & Consultancy

Software Development

Machine Learning with Python and scikit-learn

Building a model is the easy part; trusting its reported score is not. This five-day programme covers scikit-learn pipelines, honest cross-validation, the leakage patterns that inflate results, class imbalance, threshold selection, probability calibration, and the interpretation work that lets a business act on a prediction.

5 daysIn-house, online, or customized deliveryCorporate teams and professional groupsLevel: Intermediate

Overview

Practical learning for workplace transfer.

Models that disappoint in production were rarely as good as their validation score claimed. A scaler fitted before the split, a target-derived feature, a shuffle applied to time-ordered data, or an accuracy figure quoted on a dataset that is ninety-eight percent negative — each produces a number nobody can reproduce later. This course treats evaluation discipline as the core skill and the estimator as a detail. Participants build every model inside a Pipeline so preprocessing is refitted per fold, choose validation schemes that respect time and group structure, tune with nested search, set a threshold from a real cost matrix, and calibrate probabilities before they are used as risk scores.

Prerequisites

Confident Python and pandas skills, plus familiarity with descriptive statistics. No prior modelling experience is assumed.

Objectives

  • Translate a business question into a supervised learning task with a measurable target.
  • Compose preprocessing and estimators inside Pipeline and ColumnTransformer objects.
  • Identify and eliminate the data-leakage patterns that inflate validation scores.
  • Select cross-validation schemes that respect time ordering and group structure.
  • Handle class imbalance with resampling, class weights, and cost-aware thresholds.
  • Calibrate, interpret, and package a model for handover to an engineering team.

Target audience

  • Data scientists formalising an ad hoc modelling practice
  • Quantitative and risk analysts moving from statistical models to ML
  • Software engineers assigned to a predictive feature
  • Data engineers who must productionise models built by others
  • Maintenance, demand planning, and credit teams evaluating predictive use cases
  • Technical leads reviewing model results before they inform decisions

Program outline

A clear structure for the learning journey.

Program outline

Outline points are grouped in one designed block instead of being treated as separate module cards.

Module 1: Framing the Problem and the Dataset

Choosing a target definition and a prediction horizon the business can act on

Regression, classification, and ranking: matching the task to the decision

Baselines with DummyClassifier and DummyRegressor before any real model

Splitting strategy: random, stratified, grouped, or time-based

Module 2: Pipelines, ColumnTransformer, and Feature Preparation

Why fitting a scaler or imputer outside the pipeline leaks the test set

Numeric scaling, imputation, and encoding categorical variables safely

OneHot, ordinal, and target encoding and their failure modes on high cardinality

Custom transformers with FunctionTransformer and the estimator API

Module 3: Supervised Algorithms and Their Trade-offs

Linear and logistic regression with L1 and L2 regularisation

Decision trees, random forests, and where variance reduction stops helping

Gradient boosting with HistGradientBoosting and XGBoost compared

Bias-variance behaviour read from learning and validation curves

Module 4: Validation, Leakage, and Hyperparameter Search

KFold, StratifiedKFold, GroupKFold, and TimeSeriesSplit chosen by data structure

Target leakage, train-test contamination, and features unavailable at prediction time

GridSearchCV, RandomizedSearchCV, and HalvingSearchCV under a compute budget

Nested cross-validation for an unbiased estimate of tuned performance

Module 5: Imbalance, Thresholds, and Calibration

Why accuracy is meaningless on rare-event data and what to report instead

Class weights, SMOTE, and under-sampling with their side effects

Precision-recall curves, ROC AUC, and picking a threshold from a cost matrix

Probability calibration with Platt scaling, isotonic regression, and reliability curves

Module 6: Interpretation, Packaging, and Handover

Permutation importance and partial dependence read correctly

SHAP values for individual predictions and their communication limits

Persisting models with joblib, pinning versions, and recording training metadata

Monitoring drift and defining the retraining trigger before go-live

Materials provided

  • Course workbook, annotated code samples, and reference notes
  • Hands-on lab environment and starter repositories
  • Exercises, checklists, and reusable code templates
  • 4D Certificate of Completion
  • Post-course technical guidance

Training Options

Programs can be delivered in-house, online, or in a blended format depending on your team's schedule, location, and learning objectives. When an external certificate or exam is included, certification rules and fees remain under the relevant awarding body's policies, while 4D provides the training and preparation support.

Why choose 4D

4D structures this programme around one of your candidate use cases — churn, credit decisioning, demand forecasting, or failure prediction. Trainers audit the historical dataset for leakage and label quality first, then build the pipeline with your analysts and agree the operating threshold with the business owner who will act on the output, so the model leaves with a defensible decision rule attached.

Speak to 4D

Plan the right training or consultancy path for your team.

Share a few details and 4D will help route your inquiry toward corporate training, consultancy, assessment, Phoenix-enabled support, or a tailored program.