Time Series Analysis and Forecasting using Python

Learn the basics of time series analysis, including defining time series, dependence on past observations and seasonality, and forecasting future values with arima, seasonal arima, prophet, and lstm.

Contrast time series with regression by showing how a date column turns a regression problem into time series forecasting, and explore algorithms like arima, sarima, arimax, lstm, and prophet.

Learn what anomalies in time series are, how anomaly detection flags deviant points like outliers and spikes, and how ARIMA, prophet, and LSTMs handle them.

Explore the four time series components—trend, seasonality, irregularity, and cyclic—and learn how decomposition and autocorrelation reveal these patterns for forecasting.

Explore time series decomposition to break down data into trend, seasonality, irregularity, and cyclic components, then apply multiplicative or additive models in Python with statsmodels to improve future predictions.

Achieve stationarity by keeping the mean and variance constant, removing trend and seasonality through differencing or log transforms, then forecast with ARIMA, SARIMA, or exponential smoothing.

Explore tests for time series stationarity, including plots, summary statistics, and unit-root tests like the augmented Dickey-Fuller test, and interpret p-values to decide stationarity.

Learn how pre-processing and data cleaning prepare time series data for analysis and forecasting, covering missing values, duplicates, outliers, and transformation techniques like scaling, encoding, and feature engineering.

Handle missing values in data analysis and cleaning with Python. Apply deletion of rows or columns, impute with mean, median, or mode, and consider advanced algorithmic or time series imputation.

Identify and treat outliers using box plots, histograms, and scatter plots, apply IQR or quantile replacements, and decide when to remove or adjust them in time series forecasting.

Master feature scaling to normalize and standardize data for predictive modeling and exploratory data analysis, using normalization to 0 to 1 and standardization with z-scores.

Explore standardization, a feature scaling technique that converts data to z-scores using the mean and standard deviation, centering data around zero and enabling Python sklearn implementations.

Learn how to apply min-max normalization to scale heterogeneous features to a 0 to 1 range, understand feature scaling, and preview Python libraries for standardization.

Apply feature scaling with normalization and standardization in Python using scikit-learn, performing min max scaler and standard scaler via fit transform to prepare numerical data for predictive analytics.

Explore feature encoding for converting categorical variables into numerical inputs for predictive models. Learn label encoding, one hot encoding, and dummy encoding, and why they matter for X and Y variables.

Apply feature encoding to churn modeling data by handling missing values, dropping unnecessary columns, and converting categoricals with label encoding and one-hot or dummy encoding using get_dummies with drop_first.

Explore data with exploratory data analysis (EDA) to summarize characteristics, gain confidence for machine learning, and detect errors, anomalies, and data structure via visual methods.

Visualize data by presenting it in graphical form using bar chart and line charts to help end customers and managers understand insights and trends.

Identify data sources for exploration by distinguishing public and private data and gaining access. Use public data or organizational data and prepare for EDA and ETL before model building.

Handle missing values in a churn modeling dataset by exploring deletion strategies and imputation, using mean, median, mode, and forward or backward fill to prepare data for modeling.

Identify outliers in time series data using box plots, histograms, and normal distribution cues. Learn when to remove, replace, or use models not sensitive to outliers in predictive analytics.

Explore data analysis types based on data categories, focusing on univariate analysis for a single variable, especially for categorical data, with an example using customer churn at Vodafone.

Analyze bivariate analysis of two variables across numerical, categorical, or mixed data; visualize with scatter plots as age and salary and interpret correlation, covariance, and positive, negative, or zero relationships.

Explore multivariate analysis, where more than two variables—categorical and numerical—yield deeper insights, with increasing complexity as you add three or four variables.

Analyze numerical data from univariate to multivariate, using mean, percentiles, and correlation; visualize with box plots, scatter plots, and a heat map in pandas.

Explore practical data analysis with Python by performing univariate, bivariate, and numerical analyses on churn data, using pandas, seaborn, and kde, including correlation and heatmaps.

Explore how derived metrics turn existing data into new features to reveal insights using domain knowledge. Apply binning and feature encoding techniques to handle categorical data.

Convert continuous variables to categorical through feature binning, using equal width or equal frequency methods to reveal patterns, missing values, and outliers in age data.

Learn how feature encoding transforms categorical variables into numerical features for predictive analytics, covering label encoding, one-hot encoding, and dummy encoding with practical insights.

Encode categorical features with label encoding and one-hot (dummy) methods, handle missing gender values by mode, and use get dummies for geography while dropping redundant columns.

Explore time series algorithms for forecasting, including ARIMA and ARIMAX variants, ARMA, and AR models. Learn how these methods address trend, seasonality, and autocorrelation, with Prophet and LSTMs.

Learn how p, d, and q define ar, ma, arma, and arima models, test stationarity with adf or cs tests, apply differencing or log transformations, and forecast with inverse scaling.

Explore the autoregressive (ar) theory behind time series forecasting, defining ar models, pacf for selecting p, and examples of ar1 and ar2, with past values shaping demand.

Explain how moving average models compute the average of consecutive data points in a window to forecast. Show MA depends on error terms and how ARMA combines AR and MA.

Explore ARIMA concepts with air passengers data, assess stationarity, apply differencing and transformations, and forecast future passengers using ACF and PACF to choose model flavors.

Master stationarity testing for ARIMA forecasting using the augmented dickey-fuller test, apply transformations and inverse transforms, and interpret p-values and confidence levels to decide readiness.

Implement arima techniques by transforming non-stationary time series with log, double log, and differencing (including moving average variants), test for stationarity, and perform inverse transformations to obtain forecasts.

This lecture explains how to use the autocorrelation function and partial autocorrelation function to identify AR and MA orders for a log-transformed time series, with notes on grid search in production.

learn how to apply log transformations and their inverses, handle log differencing and moving averages, and compare RMSE to choose the best transformation for time series forecasting in Python.

Explore model building for time series forecasting in Python, including training and test split, log transformation and inverse back, arima forecasting, and end-to-end use cases with streamlit or flask deployment.

Explore Facebook Prophet for time series forecasting, featuring fast additive modeling with yearly, weekly, and daily seasonality, holiday effects, and robust handling of missing data and outliers in Python.

Apply Facebook Prophet to air passenger data, forecast 12 months ahead, and visualize trend and yearly seasonality with confidence bounds, using log transform and inverse back.

Use Facebook Prophet to model holiday effects by adding built-in country holidays with add_country_holidays or a custom holidays dataframe before fitting. Compare current and prior forecasts to assess improvement.

Assess forecasting performance in time series by comparing actual and predicted values across models using metrics such as MAE, MSE, RMSE, MAPE, and R2 square to identify the best model.

Explain mean squared error (MSE) as a forecast evaluation metric, present the 1/n sum of squared errors (y minus ŷ), and emphasize that MSE is different from ma.

Compute the root mean squared error (RMSE) from the mean squared error using the formula sqrt(1/n sum (y - y_hat)^2), then compare models to choose the lowest RMSE.

Compute the mean absolute percentage error (MAPE) from the absolute difference between actual and predicted values, divided by actual and multiplied by 100, for forecasting and regression analysis.

Load and visualize the load and solar generation time series with pandas and matplotlib, then handle missing values and build ARIMA forecasting after stationarity checks.

Explore demand forecasting for e-commerce data using time series models—arima, holt's winter smoothing, and facebook prophet. Split data into training and test, apply feature engineering, and compare models with rmse.

Run pre-written code in Google Colab to preprocess a weekly demand dataset, impute missing values, aggregate unit sold by week, and conduct eda with distributions and correlation before forecasting.

Decompose the weekly unit sold time series to reveal trend, seasonality, and residuals, assess volatility, and compare four week and 52 week moving averages for prep to autocorrelation modeling.