Introduction to Time Series with Python [2025]

Download all datasets attached in the lecture resources for time series with Python course and use them for coding activities. If any dataset is missing, post a question in Q&A.

Identify time series data sources such as databases, csv files, and web scraping; retrieve data with appropriate libraries and load it into a pandas data frame with a time index.

Learn to read excel files with pandas by using read_excel and specifying sheet_name, with default first sheet; inspect with ExcelFile and read multiple sheets into a dictionary of dataframes.

Combine two dataframes with identical schemas using pandas.concat to produce a single dataframe with 74,124 rows, spanning 2017 and 2018, then flatten the index with drop level.

Learn to handle date and time in time series with pandas by creating time indices, inspecting types such as timestamp and datetime, and converting with to_datetime.

Learn to work with date and time in time series using pandas. Convert to datetime, access properties, and parse formats with to_datetime for readable results.

Learn to read a dataset with pandas and convert its date column to datetime using pd.to_datetime, handle formats with dayfirst and explicit format, and index for time-series analysis.

Learn to work with pandas datetime properties, set a datetime index, and slice data between dates while generating date ranges with date_range and adding offsets with timedelta.

Identify missing values, their patterns, and their effects on analysis. Apply strategies like deletion, mean or median imputation, regression and multiple imputation, and expectation maximization or k-nearest neighbor methods.

Learn to handle missing values in time series with pandas, using read_csv in Colab, and inspect with isna, sum, and info on the CO2 dataset.

Learn how to handle missing values in time series by converting zeros and question marks to NaN, imputing data, and validating results with pandas describe.

Apply univariate imputation to handle missing data, using mean, forward fill (last observation carried forward), and backward fill (next observation carried backward), with dataset visualization.

Apply forward fill and backward fill imputation to CO2 time series data in Python, and visualize original versus imputed series.

Implement rmse scores and plot graphs to compare imputation methods such as forward fill, backward fill, and mean across CO2 and clickstream datasets.

Learn how to perform univariate imputation with scikit-learn's SimpleImputer, testing mean, median, and most_frequent strategies on CO2 and clickstream data.

Explore multivariate imputation to handle missing data by leveraging multiple variables, showing how it improves estimates over univariate methods and the synergy among dataset features.

Implement iterative imputation using regression to predict missing values across all variables. Test multiple estimators, apply the dot transform, and compare results visually.

Explore time series imputation with Python using pandas DataFrame.interpolate to fill missing values. Compare linear, quadratic, nearest, and cubic methods and see how neighboring points influence results.

Explore signal processing as a subdomain of electrical engineering, analyze time series with statsmodels, and apply moving averages to hourly to annual observations.

Implement moving averages on a sales time series using a rolling function with window sizes 3 and 5, and visualize the original data alongside the smoothed series.

Explore how window functions reduce edge artifacts and spectral leakage in time series analysis by tapering data with examples like rectangular, Hann, Hamming, and Blackman windows.

Implement window functions by computing the rolling mean on sales data with various window types, such as boxcar and Blackman, and plot the results.

Explore sdl decomposition to split a time series into trend, seasonal, and residual components using the seasonal_decompose method in Python.

Explore autocorrelation in time series, quantify how current values relate to past observations using the acf, and use patterns, such as trends and seasonality, to improve forecasting.

Compute autocorrelation with numpy, compare with pandas autocorrelation plot, read the dataset, and display results using plt.plot and plt.show.

Explore autoregression models in time series, predicting the current value from past observations using order p, guided by autocorrelation and partial autocorrelation, with estimation and forecasting.

Implement autoregressive models on sunspot data by splitting into train and test sets, fitting an AR model, and evaluating with MAE, MSE, and RMSE, then plotting predictions.

Explore autoregressive moving average models that combine autoregressive and moving average components to describe and forecast stationary time series, with parameter estimation and order selection via ACF and forecast errors.

Implement autoregressive integrated moving average models in Python using arima, train on a time series, generate predictions, compute absolute errors, and visualize results.

Explore how Fourier analysis decomposes time-domain signals into frequency components using the Fourier transform and discrete Fourier transform, with applications in signal processing, time series analysis, and communications.

Explore Fourier transform implementation for time series with Python, computing power spectra from sunspot data and visualizing the results with plots.

Explore spectral analysis filtering with the Fourier transform to analyze signals in the frequency domain, suppress noise, and isolate or remove frequency components for a cleaner time-domain signal.

Implement spectral analysis filtering using fast Fourier transform, compute power and phase spectra, and visualize them with subplots for the sunspot dataset.

Explore unsupervised methods for time series data to uncover patterns, correlations, and underlying dynamics using clustering, anomaly detection, dimensionality reduction, and density estimation.

Explore how time series anomaly detection identifies unusual data points and patterns using statistical thresholds, decomposition, autoencoders, and proximity methods across cyber security, finance, health care, and environmental data.

Identify change points in time series using change point detection, and explore cost functions, search methods, and constraints for real-time detection and anomaly detection, with connections to segmentation.

Implement anomaly detection on a time series using alibi detect spectral residual method, load the intrusion dataset, compute anomaly scores, and visualize results.

Implement change point detection on a synthetic multivariate time series using ruptures and binary segmentation. Plot and compare detected segments to identify three change points amid Gaussian noise.

Learn k-nearest neighbors with dynamic time warping to classify time series by aligning sequences of varying length, covering training and testing phases, robustness to missing data, and practical applications.

Explore the Silverkite forecasting method for accurate, interpretable time series forecasts with multiple seasonal patterns, external regressors, regression modeling, and automated uncertainty estimates.

Explore xgboost, a powerful open-source machine learning library for regression and classification that uses gradient boosting with regularization, cross-validation, and missing-value handling to boost predictive performance.

Explore the k-nearest neighbor time-series classifier with time cropping and tsfresh feature extraction, including imputing missing values, feature selection, and cross-validated performance.

Learn how to implement Silverkite for COVID time series forecasting in Python, load data with pandas, configure forecaster, evaluate with backtests, and generate 90-day forecasts with uncertainty.

Learn how to implement XGBoost for time series by building a transformer-based feature extractor, integrating with a column transformer and pipeline, and preparing train and horizon splits for future prediction.

Build and evaluate an xgboost regression model to forecast horizon values in time series, using train and test data and feature engineering.

Explore Facebook Prophet, an open source time series forecaster with additive seasonality and holiday effects. Learn to handle missing data, estimate uncertainty, and apply Prophet to inventory and marketing campaigns.

Implement time series forecasting with Facebook Prophet in Python using Google Colab, loading CSV data into pandas, fitting a Prophet model, forecasting ten years with a future dataframe, and plotting.

Learn to handle monthly data with a Prophet model in Python: import libraries, prepare the dataset, fit the model, and forecast monthly future values, then plot the results.

Learn to handle regular time gaps and missing values in time series with Python, constrain seasonality during gaps, and build a multiplicative model to forecast and plot results.

Explore solving gap in time series forecasting with python by adjusting future data, constraining axes, and plotting forecasts and seasonality to compare days in August 2018.

Explore additive and multiplicative seasonality in time series, where additive adds a constant seasonal component to the trend, and multiplicative scales with the level for forecasting.

Explore time series modeling with Prophet by comparing additive and multiplicative seasonality, fitting and forecasting data, and visualizing components in a hands-on Colab workflow.

Explore controlling seasonality in time series with Fourier order and the Prophet model, using daily data to set multiplicative yearly seasonality and forecast one year ahead.

Learn to implement a custom seasonality beyond standard cycles by modeling an 11-year sunspot cycle, and add it with a defined period and order while disabling defaults.

Explore how to model custom seasonality in time series, tune periods for hourly and quarterly patterns, and generate forecasts with future data using seasonality-aware methods.

discover how to implement conditional seasonality in time series with Prophet by building boolean condition columns for weekend and weekday, adjusting seasonality settings, and using hourly electricity usage data.

learn how to detect and handle time series outliers using Python Prophet, including data prep, plotting, and exploring seasonality and yearly patterns to improve forecasts.

Exclude outliers by using pandas boolean indexing to drop data between July 29, 2016 and September 01, 2016, then build a profit model with yearly seasonality and forecast 365 periods.

Examine the second type of outlier errors, which minimally affect the forecast but widen uncertainty; adjust the model with multiplicative seasonality and trend change points to handle future uncertainty.

Detect outliers in time series automatically using winsorization and clipping, compare extreme-value removal, and understand how stationarity shapes automated analysis.

Explore outlier detection in time series using standard deviation and percentile limits, apply winterization to cap extremes, remove points beyond 1.65 standard deviations, and visualize results with a scatter plot.

Explore how to detect outliers in time series using a rolling 300-point moving average and standard deviation, with Pandas, to compute upper and lower bounds and filter data.

Detect outliers with the final method using error standard deviation and 80% uncertainty bounds in time series forecasts, and apply a Prophet model with seasonality and regularization to avoid overfitting.

Explore downsampling and upsampling in time series with pandas, using resample to change frequency, apply daily and three-day aggregations, and fill missing values with interpolation or forward fill.

Visualize data to detect outliers and compare parametric and non-parametric methods, using box-and-whisker plots to identify outlier observations with seaborn.

Explore visualizations for time series after downsampling, including plots and boxplots, to identify potential outliers. Compute the interquartile range and use the whisker boundaries to set upper and lower fences for outlier detection.

Compare box plots and violin plots in Seaborn to visualize distributions and identify outliers. Explore percentile cues such as 25th, 50th, and 75th, and how KDE adds insight into tails.

Apply the Tukey method for outlier detection in time series by computing Q1, Q3, and IQR, setting lower and upper fences, and implementing a Python function to flag outliers.

Explore how z-score standardizes data and detects outliers from the mean in units of standard deviation, using thresholds like ±3, with notes on its parametric assumptions and non-normal data alternatives.

Implement a z-score function to detect outliers in time series data in Python by comparing values to the mean with a standard deviation threshold, and visualize results with plots.