Data Science for Business Leaders: ML Fundamentals

Explore regression and classification, the two essential model types, and how methods from linear/logistic regression to gradient boosted trees and neural networks yield numeric outputs or class probabilities.

Define a model as a tool that generalizes training data to produce outputs, noting correlation is not causation, and outline the lifecycle from population of concern to production for predictions.

Learn how parameters and hyperparameters differ, from deterministic weights in models to decision trees, and how hyperparameter optimization uses grid or random searches.

Explore the model life cycle from development to deployment, then add experience to retrain with new data, monitor performance, and replace with a better model on a regular cadence.

Explore how machine learning combines models, learning processes, and retraining to automate improvement. Explain business context, model building algorithms, end-to-end learning, and autonomous reinforcement learning.

Compare traditional statistical methods with modern machine learning, highlighting automatic feature selection, train-test splits, and the trade-off between predictive performance and explainability and transparency.

Differentiate supervised learning, which uses a target or ground truth for regression or classification, from unsupervised learning that has no target and is often exploratory.

Explore how deep learning, a specialization of machine learning powered by neural networks, handles images, video, and language. See how convolutional networks learn features and biased data mislead the model.

Explore how AutoML accelerates machine learning by automating data gathering, feature engineering, model building, testing, selection, deployment, monitoring, and documentation, while keeping human definition and interpretation in focus.

Define artificial intelligence as a model-driven system with inputs and a capability, illustrated by spam filters and self-driving cars, and contrast static and dynamic learning with data capture.

Identify common machine learning pitfalls, especially overfitting and underfitting, and learn to ensure generalization by using representative training data and proper test and validation sets.

Explore how models capture correlations to estimate outputs, differentiate regression and classification, and understand how data, ground truth, and learning processes drive evolving machine learning and artificial intelligence systems.

Part 2 teaches how to identify machine learning use cases within your organization, focusing on predictive models in production and value from structured data rather than following a predefined list.

Identify machine learning opportunities by asking the right business questions and exploring browsing, use cases, process inspection, and ML themes to guide ML solutions.

Identify cross-industry and industry-specific machine learning use cases, including attrition, targeted marketing, lead optimization, and fraud detection, and learn where to find ideas from resources like Kaggle and industry vendors.

Inspect your business processes to uncover machine learning opportunities. Analyze core and non-core activities—both physical and virtual, manual or automated—to identify where ML adds value.

Explore seven machine learning themes generalized for diverse business contexts, and learn when and how machine learning might fit into a particular business setting.

Learn to allocate limited resources with machine learning by ranking opportunities, using regression and classification with well calibrated probabilities to optimize marketing and prospecting.

Analyze how machine learning augments human decisions by considering more inputs, including promotions and local events, to predict store foot traffic and optimize staffing and planning.

Explore machine learning theme five: analyze activities at scale to identify repetitive tasks ripe for ML, and leverage models in production to improve estimates and predictions.

Explore how to predict business-critical events with machine learning through binary classification, using sensors and data to forecast failures within useful time windows.

Explore how machine learning predicts people across contexts like customers or patients. Consider ethical concerns, fairness, and accountability—would you stand behind your model's decisions in practice.

Explore how AI capabilities enable automation and improved decision making, focusing on chat bots, NLP and NLU, intents, sentiment analysis, data needs, and dynamic AI risks.

Establish a business hypothesis to define value and how to achieve it. Express it as X leads to Y, for example improving response rate by 1% yields 100 customers.

Develop a high-level roi estimate and a rough measurement framework to guide feasibility decisions, socializing key assumptions and exploring likelihood and value models for customer prospects.

Lock down the model's population and target to align with the business process. Scrub data before modeling and use the expected value, likelihood times value, to rank prospects.

Learn how data targets and ground truth determine model quality. Use test and learn and proxies for cold start scenarios when historical labels are unavailable.

Assess data feasibility by identifying informative features among columns and considering third-party data, then determine required rows through learning curves, starting small and scaling with predictive power.

Ensure data is available going forward and consistent, with stable features and pipelines ready for production, and test readiness with a one-time data pull.

Evaluate the feasibility and performance of a built model by analyzing fit, errors, and feature importance across subpopulations, and consider target leakage, explainability, and the go versus no go decision.

Measure the utility of a regression model from a business perspective using simulation. Establish baselines, plan data splits, and consider seasonality and time window effects to evaluate business process performance.

Explore how to measure regression performance with mean absolute error (MAE). Compute absolute errors from model predictions and interpret MAE in dollars for business insights.

Evaluate a regression-based method for ranking prospects by simulating top-100 calls, compare mean absolute error to the baseline, and estimate ROI from historical outcomes and probabilistic results.

Apply faithful simulations of repeated business processes, such as outbound marketing, accounting for non-independent events and changing customer populations, and run many simulations to produce robust, aggregated estimates over time.

Use the historical median as the baseline when no prior estimates exist. Compare the model’s mean absolute error to this baseline, and consider skewness that shifts mean versus median.

Assess regression performance using mean absolute error before and after improvements. Use repeated simulations to estimate value and return on investment, and apply guardrails for go/no-go decisions.

Establish a baseline, use simulation, and analyze a binary classifier through the confusion matrix, accounting for type one and type two errors, prediction probabilities, and accuracy weaknesses.

Examine accuracy limitations in classification, reveal how class imbalance and majority class bias can inflate metrics, and suggest area under the curve as a more reliable measure of predictive power.

Compare two cancer classifiers via confusion matrices, focusing on outcome value over accuracy. Prioritize classifier two for its 90% sensitivity, accepting some false positives to minimize missed cancers.

Learn how calibrated models output probability of class membership to rank prospects by expected value, enabling budget-driven mail targeting and threshold-aware decision making.

Explore performance measurement for classifiers by using probability of class membership and expected value to rank outcomes, set business-driven thresholds, and compare baselines through simulation.

Learn to validate a binary classifier with a calibration plot of predicted probabilities versus actual outcomes, using deciles to assess calibration and model fidelity.

Apply experimental design to machine learning initiatives by using control groups, A/B testing, and design of experiments to diagnose degradation, validate hypotheses, and optimize data collection and decision thresholds.

Evaluate the business viability of machine learning initiatives by assessing pre and post modeling feasibility, ROI, data readiness, and costs, then use simulation to inform go/no-go decisions and model maintenance.

Assess your organization's context for machine learning to decide whether to build or outsource. Explore how mlops, talent decisions, data considerations, and cloud implications shape ml capability.

Evaluate business software with AI or ML by scrutinizing learning process, model inputs, and updates; ask about outcomes, training data, model specificity, and data flow to separate hype from value.

Explore deployment configurations for machine learning systems across desktop software, enterprise solutions, and third-party hosted models, and assess data flows, telemetry, and risk.

Evaluate machine learning software by asking about transparency, data sources, and where learning occurs; assess dependencies, customization, governance, risk, and ROI to avoid inefficiencies.

Learn how to plan, manage, and govern ML consulting engagements with emphasis on data readiness, information privacy and security, specialization, integration, validation, maintenance, and documentation.

Explore infrastructure for ml projects and the role of mlops in data gathering, feature engineering, model building, testing, deployment, and monitoring to improve operational efficiency.

Build robust data infrastructure by preserving timestamped data with traceability, linking across sources at the lowest level, and preserving ground truth for predictors and outcomes while handling missing data.

Understand the two compute stages for machine learning—development and inference—and how memory and gpus or tpus drive performance, enable api hosting, containerization, load balancing, and scalable deployment.

Explore how cloud computing powers machine learning infrastructure, highlighting scalability, cheap storage, fast deployment, and built-in algorithms, while noting trade-offs like vendor lock-in and support.

Explore how AutoML contributes to your machine learning infrastructure across data gathering, model development, deployment, monitoring, and reporting. Compare Datarobot with cloud options to avoid dual infrastructures.

Sprint to a two-week data collection and build a prototype model. Decide centralized or decentralized data science teams, address risk management and governance, and document assets for ML competency.

Explore opportunities for today and the future by applying machine learning, running use case workshops, and utilizing probability and multiple models to guide decisions.

Explore AutoML as automation across the machine learning process from development to deployment, while acknowledging that model developers and data scientists remain essential for feature engineering, interpretability, and explainability.

Conclude the course by equipping business leaders with practical, non-technical machine learning fundamentals to inform strategy, measure impact, consider purchasing established competencies and strategy, and pursue organization training and feedback.