RA: Retail Customer Analytics and Trade Area Modeling.

Learn to install Anaconda through a simple, step-by-step process and discover which applications run inside the environment.

Explore Spyder overview, its console and script editor, and compare it with notebooks for Python coding, project setup, and file management in a supply chain data science workspace.

Explore how to use Jupyter Notebook and Spyder inside Anaconda to write and run Python code, view instant outputs in cells, and switch between environments.

Explore Python libraries such as pandas, numpy, and matplotlib for data manipulation, visualization, and forecasting workflows, and import them in Jupyter notebooks or Spyder within an Anaconda environment.

Learn Python data types and structures for retail analytics, including numbers, strings, datetime, categorical data, booleans, lists, arrays, dictionaries, and tuples, with parsing of missing values.

Explore how Python uses parentheses for function calls and assignment with equals, distinguish objects from strings, and data frames resemble Excel tables with rows as observations and columns as features.

Explore basic Python arithmetic in a Jupyter Notebook, creating variables like addition and multiplication, building a prime numbers list, and using zero-based indexing to access items.

Explore how dictionaries map keys to values, create dictionary examples, and retrieve keys, values, or specific elements using dot notation and indexing.

Explore numpy arrays in Python and contrast them with lists and dictionaries. Learn that arrays are memory-efficient and typically one-dimensional, with elementwise arithmetic.

Import data in Python using Pandas in a Jupyter notebook and load online_retail_two.csv from Kaggle, then explore with head, tail, shape, and describe to analyze rows, columns, and statistics.

Learn to subset a data frame using loc and iloc, selecting rows and columns, and add a new price in europe column by applying a hypothetical conversion to usd.

Explore how to implement and test conditions in data analysis, using comparisons like bigger than, smaller than, bigger or equal, and combining multiple criteria in for loops and if statements.

Create a Python function named status using def, if, elif, and else to classify a person's age as child, teenager, or adult, with indentation and optional print or return.

Learn how to apply a Python map function to a list, creating a map object and converting it to a status list to see each element's result.

Learn how for loops function as an essential iteration tool, using a class list to print each age and compare the approach with the map function.

Demonstrates using a for loop to apply a status function to an age list, store results in a status list, and compare with map for data frame preparation.

Define a function to flag United Kingdom entries in a retail data frame, then map country values to a new UK column with true or false results.

Apply a for loop to a data frame, reproducing mapping output for the first ten rows and updating a column with a function.

Master Python basics—from lists and dictionaries to pandas read_csv, iloc, and loc—for dataset subsetting, and use functions, map, and for loops to support retail analytics and trade area modeling.

Practice a Python assignment on the cars dataset to count rows and columns, compute horsepower metrics, and create a price-based category (budget, suitable, expensive) with a for loop.

Analyze the cars dataset to explore horsepower and price, compute the shape for rows and columns, and identify the Porsche 911 Gt2 as the most expensive.

Rename the dataset's name column to car name, create a car pricing subset with name and price, and classify prices into budget, suitable, and expensive.

Master data manipulation in Python for retail analytics, unifying data with pd.concat, merge, and join. Create new columns, group data, and index for subsetting and summarizing across classes and grades.

Impute and drop unnecessary columns, convert the invoice date from object to datetime, create a date column, and enable time-based analyses like recency and purchasing frequency in retail analytics.

Learn how to filter data in Python by creating subsets from the retail data, using equals and isin for multiple countries, and applying date filters like date >= 2011.

Filter between multiple dates, such as August to December, and impute values by targeting a column like country or stock code to update Eastern Island to Eastern Ireland.

Organize data with row and column indexes to quickly filter by date or country, and explore multi-level indexing like country and date, using set index and reset index.

Master three core data manipulations—pivot, melt, and group by—along with aggregate and time series reshaping to support forecasting and flexible summaries like mean, standard deviation, and percentiles.

Use the aggregate function to apply mean and median to quantity and price, then slice by country level Australia and price 0.85, and remove the index for a normal table.

Learn to use aggregate to keep the identity of every column, including the country column, intact while computing mean and median for quantity and price, and reset or preserve index.

Master pivot tables to reshape supply chain data. Create a data frame with country, date, and quantity; pivot by country, and build a United Kingdom time series.

Apply an aggregate function in a pivot table to sum quantity by date and country, and compute mean price for these dimensions, producing a two-level pivot with quantity and price.

Join level zero and level one with an underscore to create a single-level column, then melt the data to long format, using date as the id and measure as the variable.

Learn to perform a left join between sales and department tables with a common key using pd.merge, preserving all left-table rows while attaching matching department attributes when available.

Practice joining two data frames in Pandas to combine names, designations, and ages. Compare inner join, full outer join, and left join workflows as you merge the tables.

Learn data manipulation for retail analytics, including conversions, cleaning with dropna and drop, groupby and aggregate, pivot tables and melt, indexing for dates, joins, and the New York Airlines assignment.

Analyze the New York Flights 2013 dataset to answer questions about popular destinations, busiest months, punctuality, delays, and aircraft capacity by merging data frames and using group by.

Import pandas and load flights, planes, airlines, and airports data with read_csv, then inspect columns like year, month, departure time, and departure delay.

Count flights from New York by destination, sort to identify the top city, then left join with airports to map names, revealing Chicago O'Hare International as the top destination.

Explore how to identify the busiest month by counting monthly flights, and assess airline punctuality by computing total and mean delays per carrier to pick the most punctual airline.

Demonstrate calculating the longest destination airtime by grouping flights by origin and destination, computing the mean airtime, and using a left join with airport data to confirm.

Identify the airline with the highest seating capacity by linking flights to planes, summing seats per tail number, and aggregating by carrier; reveal the most used model and its manufacturer.

Explore trade area modeling to choose store locations using the half and hough models in python, focusing on attractiveness, distance, drive time, proximity, parking, and competition.

Assess trade area modeling to determine optimal store location by weighing cost, accessibility, footfall, and nearby competition, and using data on population, destination attractions, traffic, and business mix.

Compare drive time and data driven trade area models to assess store location viability. Use radius, obstacles, and zip code insights to balance competition, footfall, and customer origins.

Apply the Hough model to estimate store choice probabilities using attractiveness and distance, with a beta coefficient, and enrich with community spending and income data.

Explore how trade area modeling differs for convenience and destination stores, including calibration with the alpha coefficient, the roles of distance, store attractiveness, customer types, and seasonal effects.

Explore the Huff model for trade area analysis, computing the probability a consumer shops at a store from its attractiveness and distance to the alpha power, then derive market share.

Explore a gravity model approach to retail site selection by calculating distance from communities to stores, measuring attractiveness, estimating purchase probability, and deriving expected sales and market share.

Explore scaling households, grocery expenditure, and attractiveness attributes for the half gravity model in Excel and Python, comparing min-max and z-score methods, then apply weighting.

Scale attributes with min-max normalization to avoid bias, then compute store attractiveness and Huff probabilities using distance with alpha, and estimate expected sales to guide trade-area opening decisions.

Apply the Huff model in Python to guide store placement using a distance matrix and community data. Measure attractiveness with size, households, income, traffic, accessibility, and competition, implemented via dictionaries.

Learn to read the python trade Excel file with three sheets using pandas and numpy, organize distance, census, and stores data, and build a complete store–community key.

Scale store attractiveness with a minimax method, then compute the upper term as attractiveness over distance squared and derive store probabilities for trade-area modeling.

Learn to compute expected store sales by multiplying key probabilities by total expenditure, building per-key and per-store totals from census data, and identify the brand with the highest expected sales.

An assignment to select a store location in Los Angeles County using a distance matrix, households, grocery expenditures, and attractiveness attributes; implement in Python with looping and dictionary filtering.

Apply pandas and numpy to build distance, attractiveness, and census data frames, scale attractiveness with MinMaxScaler, compute PIGS and expected sales, and identify Lakeview Terrace as the best store prospect.

Explore RFM analysis (recency, frequency, and monetary value) for customer segmentation and lifetime value prediction, and compare statistical grouping with K means clustering using the elbow spray and silhouette score.

Explore RFM segmentation to classify customers by recency, frequency, and monetary value, identify loyal and heavy spenders, and tailor marketing campaigns with traditional and k-means approaches.

Explore RFM analysis to segment customers by recency, frequency, and monetary value, identifying core, loyal, luxury, challenge, and personalized segments, and apply ranking and k-means methods for targeted retention.

Clean and prepare retail data with pandas and NumPy, convert invoice dates to date objects, derive each customer's last purchase date, and compute recency using the dataset max date.

Convert recency to integer days, compute per-customer frequency from invoice counts, and calculate average monetary value from revenue to prepare an RFM profile per customer.

Rank customers by recency, frequency, and monetary value to identify which customers are most engaged; learn to merge these rankings to analyze segments and drive targeted strategies.

Plot a count plot of recency, frequency, and monetary value to define rfm-based customer segments and guide targeted marketing strategies.

Apply the k-means clustering algorithm to segment customers by recency, frequency, and monetary value, exploring centroid counts and convergence to balance under- and over-segmentation.

Visualize centroids on retail customer data with a pair plot, revealing segregation by recency, frequency, and monetary value, and determine the optimal number of centroids with the elbow.

Apply the elbow spray method to identify the optimal number of centroids by locating the elbow in the SSE plot, typically around three clusters.

Apply k-means to mall customer data. Use eda plots of age, income, and spending score; scale features; evaluate elbow and silhouette scores to define clusters and identify high-spending group.

Analyzes a customer dataset using numpy, pandas, seaborn, and matplotlib; applies minmax scaling and k-means with elbow and silhouette analysis to reveal clusters, including a high-income, low-spending segment around 41.

Explore estimating customer lifetime value over a year using RFM attributes (recency, frequency, monetary value) and demographic factors, building regression or classification models in Python to forecast next year’s spend.

Use RFM features to compute customer lifetime value, segment into high/medium/low value with K-means, and predict LTV using a decision tree model and multinomial logistic regression for targeted marketing.

Compute customer lifetime value from statistics by cleaning data and computing RFM. Split recency, frequency, and monetary into columns, map values to reverse rankings, and derive the overall RFM score.

Compute the overall RFM score by converting strings to int64 and summing recency, frequency, and monetary groups, then derive lifetime value (LTV) from revenue (quantity times price) for customers.

This lecture handles LTV outliers by scaling and trimming to the 99th percentile, reducing noise, then uses k-means to classify LTV into low, mid, and high.

Join the RFM and LTV data, remove outliers, select key features such as recency, frequency, monetary value, and group scores, then get dummies for groups to model.

The lecture compares logistic regression, multinomial logistic regression, and a decision tree classifier without tuning, using repeated stratified k-fold cross-validation on about 5800 observations and achieving about 85% accuracy.

Tune a decision tree and a random forest using randomized search cross-validation. Compare max depth, min samples leaf, and splitting criteria (gini or entropy) with five times cross-validation.

Tune the model grid with cross-validated randomized search to optimize parameters, compare rf and tree variants, and generate predictions with actual versus predicted data.

Apply RFM analysis to UK retail data, map recency, frequency, and monetary value, cluster with k-means, and build random forest, logistic regression, and decision-tree models to assess customer accuracy.

Import UK retailer data, split recency, frequency, and monetary value into RFM groups, compute an overall lifetime value score, remove outliers, and apply k-means clustering.

Explore how logistic regression identifies features driving churn and how calls to support and voice plans affect churn probability and classification accuracy using log odds and odds ratios.

Churn prediction is framed as a binary classification problem and, using logistic regression, helps identify customers likely to churn and the factors driving churn for subscription-based businesses.

Explore how logistic regression extends linear regression to predict churn as a probability, using odds, odds ratio, and a telecom Kaggle dataset to interpret feature effects.

Explore probability, odds, and odds ratio with simple color examples comparing red M&M and red Skittles, and relate these concepts to the Fisher test for sample comparisons.

Explore logistic regression, extending linear regression to binary outcomes by transforming AX+B with the logistic function to produce probabilities from 0 to 1, using odds and log odds.

Import data in notebook and load train and test CSVs. Identify and convert categorical variables like state, area code, international plan, and churn to category types for one hot encoding.

Import data, verify non-null variables, and encode binary features like international plan and churn; visualize churn impact with box plots of total minutes and calls to reveal patterns.

Analyzes churn rate by state using group by and sum, sorts states by churn, and uses histograms to compare churn with customer service calls and evening minutes.

Prepare data for modelling by applying correlation on continuous variables, performing one-hot encoding, and creating train-test splits using the stats library for logistic and linear regression.

Build and interpret a logistic regression using statsmodels, evaluating coefficients and p-values to determine feature effects on churn, noting international plan significance and potential detriment from area code and state.

Evaluate model accuracy and explain precision and recall. Use true positives, false positives, and false negatives from the confusion matrix to discuss trade-offs between recall and precision.

Explore log odds and odds ratios in a logistic regression to see how account length and international and voicemail plans influence churn. Interpret coefficients; exponentiate log odds for odds ratios.

Learn how logistic regression extends linear regression, how to manually add an intercept with statsmodels, and how log likelihood, pseudo r-squared, and odds ratios inform churn analysis.

Master logistic regression with smf formulas and logit to model churn and interactions. Assess feature effects via coefficients and p-values using training data.

Develop an interaction model in logistic regression using Patsy design matrices, integrate scikit-learn tools for model selection, and drop area code and state to optimize multiplicative interactions.

Demonstrate fitting the interaction model with ridge-penalized logistic regression, using grid search cross-validation to tune the alpha parameter, and assess accuracy on train and test data.

Explore lasso and ridge regression for retail analytics, using interaction terms and metrics like accuracy, confusion matrix, precision, and recall, with cross-validated alpha to identify the best model.

Extract coefficients from the best lasso estimator, map them to interaction terms and features, and interpret their impact on churn probability and odds in logistic regression.

Describe and preprocess a Kaggle telecom churn dataset with 21 features, then use logistic regression to assess accuracy, precision, recall, and false positives, comparing all factors vs. significant ones.

Delve into market basket analysis and association rules to reveal which products pair up, quantify support and confidence, and enable recommendations, bundles, and streamlined one-click shopping.

Explore market basket analysis to uncover product associations and customer recommendations. Learn key metrics: support, confidence, lift—and the a priori algorithm, plus data preparation and practical retail applications.

Install and import the extend package and core libraries, load the retail clean data, compute revenue, and set up for apriori-based association rules in market basket analysis.

Visualize a top ten dot count plot of item descriptions with rotated labels, and note wholesale orders average 21 items, median 15, with outliers visible in a box plot.

Prepare data for market basket analysis by turning invoices into a binary matrix with items as columns and presence indicators, using group by, unstack, and zero filling for apriori.

Encode baskets with one-hot encoding, convert to Apriori antecedents and consequences, and derive frequent itemsets and association rules using a minimum 1% support and confidence ranking.

Identify slow moving items with percentile-based abc classification and association rules, then offer promotions and recommendations on a serverless e-commerce platform to boost slow item sales.

Refine market basket analysis by raising minimum support to 9%, parsing antecedents and consequences as strings, and using eight data cuts to identify slow-moving item bundles for promotions.

Explore recommendation systems in retail analytics, comparing collaborative filtering and content-based approaches, with emphasis on the surprise SVD algorithm and market basket concepts.

Explore how recommendation systems power e-commerce, streaming, and social platforms by using collaborative filtering and content-based filtering to match users with items based on behavior and attributes.

Explore collaborative filtering versus content-based filtering for retail analytics, comparing user-based and item-based approaches, illustrated with rating matrices, sparsity, and singular value decomposition inspired by Netflix's challenge.

Prepare a retail recommender model in a notebook using scikit-learn and Surprise, loading the Home Improvement Dataset with user, item, and rating, and evaluate with cross-validated SVD.

Train an svd-based model on the full dataset after cross-validated evaluation, achieving a mean rmse near one rating, then prepare the model for serving predictions on the training set.

We trained the model on the full dataset to predict a given customer's item ratings, despite sparse data. We sample 1% of items and rank top recommendations by predicted ratings.

Compare singular value decomposition and SVD++ using cross-validation and RMSE to pick the best model, retrain on full data, and generate top recommendations for three customers.