Master Time Series Analysis and Forecasting with Python 2026

Master time series analysis and forecasting in Python by exploring data, identifying seasonality, and applying models from exponential smoothing to ARIMA, while practicing with real data and a capstone project.

Learn to explore retail data by counting stores, subsampling ten stores with a fixed random state, and compute mean sales per store, then identify each store's all-time high day.

Standardize weekday sales with z-scores to enable apples-to-apples store comparisons. Group by store and day of week and analyze mean standardized sales to reveal Sunday seasonality and other patterns.

Remove day-of-week seven observations for store 353 to clean the data, then standardize sales by store and analyze intra-week seasonality with a visual intro-week plot.

Analyze the impact of promotions by comparing average sales on promotion days versus non promotion days, then compute per store promo uplift and identify top and bottom performers.

Learn seasonal decomposition of time series into trend, seasonality, and error terms. Distinguish additive from multiplicative seasonality using visualization and model performance in Python.

Visualize seasonality in time series using month plots and quarter plots from Statsmodels, compare actual monthly values to their averages, and explore resampled monthly data.

Explore autocorrelation by comparing a time series with its lagged values to reveal how past information helps predict the future.

Explore time series forecasting of stock prices by modeling errors with relevant regressors, assessing data relevance, and understanding external factors like events, economic conditions, and sentiment.

Conduct exploratory data analysis on weekly sales using acf and pacf with 60 lags and perform multiplicative seasonal decomposition with a 52-week period to reveal trend, seasonality, and residuals.

Explore a Telco Wave case study to forecast weekly customer complaints, analyze data and hidden patterns, and align the workforce with customer needs to reduce complaints and boost profits.

Analyze seasonality in weekly data, resample to monthly with month end, and assess multiplicative versus additive seasonal models using Holt-Winters, guided by autocorrelation insights.

Apply double exponential smoothing to capture both level and trend in sales data. Use alpha and beta to weight recent data and trend changes, noting no seasonality.

Learn to measure forecast accuracy with mean absolute error, RMSE, and MAPE. Explore their tradeoffs, including interpretability, outlier handling, and when to compute them in Python.

Analyze model outcomes by computing RMSE, MAE, and MAPE using sklearn.metrics, comparing true versus predicted values, and formatting results in a reusable model assessment function for time series evaluation.

Load bitcoin price data and prepare daily data with a date index and adjusted close. Fit a Holt-Winters model, test seasonal periods of 7 and 365, and discuss crypto volatility.

Study the capstone air miles time series by setting monthly frequency, visualizing data, forecasting the next 12 months with holt-winters, and evaluating errors with MAE, RMSE, and MAPE.

Learn to prepare monthly time series data in Python, visualize trends and seasonality, perform multiplicative seasonal decomposition, assess ACF and PACF, and set up training and test sets.

Complete task three: set the last 12 months as the test set and the training data. Build the Holt-Winters time series model with exponential smoothing and multiplicative seasonality.

Learn to forecast 12 periods with an exponential smoothing (holt-winters) model, assess accuracy with metrics, visualize training, test, and future predictions, and explore EDA and frequency as building blocks.

Forecast daily revenues for a chocolate shop using past sales data to uncover patterns, seasonal trends, and holidays, then plan inventory and staffing with ARIMA models and external regressors.

Explore the autoregressive component of Arima, where the past values and lagged data predict the future, with examples like coffee consumption and chocolate revenues.

Explore the integrated component of ARIMA by applying differencing to convert nonstationary data into stationary signals, and learn how the augmented Dickey-Fuller test checks stationarity to improve predictions.

Assess stationarity with the augmented Dickey-Fuller test, interpret the p-value against 0.05 for the null of a unit root, and apply differencing to obtain a stationary time series.

Explore ARIMA in practice: use past values t-1, t-2, t-3, a constant, and forecasting errors with differencing and a rolling forecasting approach, building toward sarima.

Explore how SARIMA combines ARIMA components, seasonal lags at t-7 and t-14, and past forecast errors to generate a final forecast.

Explore building a sarimax-style forecast by combining non-seasonal arima components, seasonal terms, and exogenous variables like discount and coupon, resulting in a six-component forecast; then master cross-validation and parameter tuning.

Tune parameters to maximize forecast accuracy by systematically testing autoregressive models with different lags, using a reusable template and automated error tracking in Python.

Explore parameter tuning for time series models by building an eight-combination parameter grid, applying time-series cross-validation with RMSE, and optimizing forecasting performance.

Explore parameter tuning results for a time series model by transforming the RMSE grid into a data frame, identifying the best params, and preparing for future forecasting with lagged features.

Isolate the target and features, load and format future regressors, apply best parameters, adjust units by multiplying by 100, and prepare to predict future discount and coupon rates.

Master modern time series forecasting with Prophet and Silver Kite. Set up the Python environment, model seasonality and holidays, and tune parameters for accurate forecasts.

Forecast bike rental demand with Prophet and Python through a hands-on bike-sharing case study. Prepare data, perform eda, engineer features, tune parameters, and visualize insights.

Explore structural time series by decomposing data into trend, seasonality, exogenous regressors, and an error term, then use Prophet with dynamic holidays and built-in cross-validation for practical forecasting.

Analyze bike sharing demand using a rich Washington DC dataset from 2011–2012, exploring temperature, weather, date, and city events to forecast demand with practical, profit-oriented models.

Renames variables and formats dates, converts the weather situation into categorical dummies with get_dummies and drop first, and drops unused columns to streamline the data frame for time series analysis.

Perform exploratory data analysis in Python using a code template to set the date stamp as the index. Observe trend and multiplicative seasonality, plus spikes and autocorrelation and partial autocorrelation.

Model dynamic holidays like Valentine's Day by capturing event-driven demand with a lower window of minus three days and an upper window of one day after, using profit.

Explore Prophet model parameters, including multiple seasonalities (yearly, weekly, daily), multiplicative or additive seasonality, holidays, and priors for seasonality, holidays, and trend to balance overfitting and underfitting.

Identify spikes in residuals, inspect metrics, and fix the data by replacing oct 29–30, 2012 with oct 28’s values to stabilize rmse and map before feature engineering.

Explore feature engineering for time series by creating lagged temperature features (lags 1, 3, 5, 7), assessing their correlation with demand, and selecting a single lag to improve model performance.

Build a parameter tuning pipeline in Python for time series models using cross validation and RMSE as the error metric.

Create a pandas dataframe of all parameters, evaluate tuning results, and select the best additive seasonality configuration to guide future predictions.

Learn to build a Python-based time series forecasting workflow, preparing data, engineering features like holidays and lag, tuning the model, predicting the future, and visualizing results.

Build a tuned Prophet model by cleaning the training data, removing NaNs, and training only the last 30 days using the best parameters to forecast future values.

Visualize time series forecasts with Prophet in Python using ChatGPT-assisted code. Explore components, holidays, weekly and yearly seasonality, and regressors with interactive Plotly visuals and actual versus forecast.

Assess Prophet's flexibility, built-in visualization and cross-validation, and its handling of non-linearity and multicollinearity, while noting slower optimization and poor short-term dynamics for sudden changes.