Machine Learning for Business Forecasting

Section 1: Framing Machine Learning Problems from Business Needs

• How forecasting, segmentation and classification answer different business questions
• Set the goal, the metric and the decision the model's output will drive
• The data you need and the data problems organisations commonly face
• Lab: set up Google Colab and load sample sales data with pandas

Section 2: Lab: Exploring Time Series Data

• Resample data to daily, weekly and monthly levels
• Separate trend and seasonality with charts and decomposition
• The effect of Thai public holidays, festivals and promotions on sales
• Handle missing dates, outliers and slow-moving products

Section 3: Baselines and Forecast Evaluation

• Naive, seasonal naive and moving average baselines
• Choosing a metric: MAE, RMSE, MAPE and WAPE, and their limits
• Rolling-origin backtesting instead of random splits
• Lab: set a baseline that later models must beat

Section 4: Lab: Sales Forecasting with Prophet

• How Prophet is built: trend, seasonality and holidays
• Add Thai holidays and promotions as extra regressors
• Read the uncertainty interval of the forecast
• Lab: forecast 12 weeks of sales and compare with the baseline

Section 5: Lab: Forecasting with Gradient Boosting

• Build features from lags, rolling averages and the calendar
• Train LightGBM or HistGradientBoosting from scikit-learn
• Time series cross-validation to avoid peeking into the future
• Lab: compare accuracy with Prophet and the baseline

Section 6: Forecasting Many Products and Using Forecasts for Planning

• Forecast many branches and products with a single model
• Roll results up to category and company level so the numbers agree
• Turn forecasts and uncertainty ranges into safety stock
• Lab: summarise next month's ordering plan from the forecast

Section 7: Lab: Customer Analysis with RFM

• Calculate recency, frequency and monetary value from purchase history
• Score and group customers with RFM
• Name segments in terms marketing understands, such as loyal and drifting customers
• Lab: summarise customer counts and sales for each segment

Section 8: Lab: Customer Segmentation with K-Means

• Choose features and scale data before clustering
• Choose the number of clusters with the elbow method and silhouette score, alongside business judgement
• Describe each segment with tables and charts
• Lab: design a campaign that fits each segment

Section 9: Defining Churn and Preparing the Data

• Defining churn for subscription businesses compared with retail
• Set a cut-off date to build labels and features correctly
• What data leakage is, with examples that make a model look better than it is
• Lab: build a customer feature table from usage data

Section 10: Lab: Churn Prediction Models

• Logistic regression, random forest and gradient boosting
• Handle imbalanced data with class weights
• Measure with ROC-AUC, precision, recall and lift
• Lab: pick the best model with cross-validation

Section 11: Explaining Results and Deciding by Cost

• What feature importance and SHAP tell you, and what to watch out for
• Choose a threshold by weighing campaign cost against the value of retained customers
• Rank high-risk customers for the sales or CRM team to contact
• Monitor the model as customer behaviour changes and set a retraining cycle

Section 12: Workshop: Capstone Presentation to the Business

• Choose a problem: sales forecasting, customer segmentation or churn prediction
• Work end to end from data preparation and modelling to evaluation against a baseline
• Export results to CSV or Google Sheets for the team to use
• Present the results in business language and review them together