Agglomerative Hierarchical Clustering

Agglomerative Hierarchical Clustering (AHC)

• AHC starts with each data point as its own cluster.
• The algorithm merges the two most similar clusters iteratively.
• The process repeats until all points are in a single cluster.

Determining Similarity Between Clusters

• Criteria to determine similarity include Euclidean distance and Manhattan distance.
• The choice of criteria depends on data characteristics and application goals.

Dendrogram Interpretation

• Based on the provided dendrogram, clusters can be assessed for similarity by their linkage distance.
• The shorter the vertical lines connecting clusters, the more similar they are.

Dendrogram and Linkage Methods

• Different linkage methods (e.g., single linkage vs. complete linkage) change dendrogram structure.
• Single linkage considers the shortest distance between points in clusters.
• Complete linkage considers the longest distance between points in clusters.

Practical Applications of Hierarchical Clustering

• Applications include customer segmentation and document clustering.
• The method benefits real-world scenarios by revealing inherent data structures.

K-Means Clustering

• K-means algorithm aims to determine optimal grouping based on height and width in the input dataset.
• The algorithm typically minimizes the within-cluster variance
• The output is a grouping of the individuals

Targeted Marketing Strategies

• Clusters can be applied to develop targeted marketing strategies for different body types.
• Marketing approaches can be tailored to specific clusters.

Challenges and Limitations

• Potential challenges arise when using only height and width as features for clustering.
• The challenges may be due to the non-uniqueness of height and weight combination.

Multi-Item Association Rules

• Association rule mining identifies relationships between items in a dataset.
• Rules are typically expressed in the form of "If A, then B".

Usefulness for Marketing Activities

• Association rules can be useful for marketing activities such as poster design.
• They are useful for menu customization and setting pricing strategies.
• Association rules are also useful for promotional campaigns.
• For example, discovering that customers who buy avocados, tortilla chips, and salsa also tend to buy guacamole mix and lime may prompt a grocery store to place these items near each other.
• Customers would then be more likely to purchase all mentioned ingredients at once.

Usefulness for Logistics/Warehousing

• Association rules can be useful for logistics and warehousing management.
• It enables controlling inventory levels while maintaining appropriate supply and demand.
• The logistics example might involve identifying the associations of product that a customer is very likely to purchase together.
• The business can prepare its supply chain and plan the deliveries for all items at once.

Poster Design

• Design a poster for a grocery store that promotes the associations found in the dataset.
• The poster should recommend common product combinations with discounted prices.

Shelf Arrangement Layout

• Design a shelf arrangement layout based on the dataset patterns.
• An example is placing items frequently bought together near each other.

Promotional Pricing Strategy

• Set a promotional pricing strategy based on association rules.
• An example would involve discounting items with strong associations.

Extraction of Valuable Information from Clustering Data

• Extract valuable similarities from a clustering structure and use it for real-world marketing purpose.
• Extract valuable dissimilarities from a clustering structure and use it for real-world marketing purpose.
• Valuable similarities are found when high ratings are found across items in the same cluster.
• Dissimilarities are found when ratings differ across clusters.

Valuable Similarities

• Identify which categories in the clustering table exhibit similarities
• Use these similarities for product placement.

Valuable Dissimilarities

• Identify categories that have low ratings

Real Marketing Purposes

• Use valuable similarities
• Product placement
• Marketing deals

Model Interpretability

• A model's interpretability is crucial for explaining decisions to non-technical stakeholders.
• The transparency determines the acceptance of the model.

Dealing with Noisy Data

• When handling complex datasets with a lot of noise, the models that apply tree-based methods are often much more robust.
• Tree-based methods benefit from the wisdom of crowds.

Quick Classification Task

• For a quick and simple classification task on a low-resource system, a basic technique is most suitable
• Basic techniques are e.g. linear models

Influence of Individual Features

• Analyzing the influence of individual features on the outcome is an important step for model development.
• Understanding feature importance enables better models and insights

Model for Facial Recognition

• When a model is developed for facial recognition in a mobile app-deep learning is a practical approach.
• Deep learning can be used for more complex recognition tasks

Calculating Support

• Support is the fraction of transactions that contain all items in the rule.
• Support reveals how frequently the itemset occurs in the dataset.

Calculating Confidence

• Confidence is the proportion of transactions containing X that also contain Y.
• It measures the reliability of the association rule.

Calculating Lift

• Lift measures how much more likely Y is purchased when X is purchased.
• It assesses the strength of the association between X and Y.

Individual Entropy

• Calculate the individual entropy at each node and leaf of the tree.
• Entropy measures the impurity or uncertainty in a set of instances.

Weighted Average Entropy

• Compute the weighted average entropy at each level of the tree.
• Weighted average entropy considers the proportion of instances in each branch.

Information Gain

• Calculate the Information Gain (IG) for each split.
• IG represents the reduction in entropy achieved by splitting on an attribute.

Interpretation of Information Gain

• Interpret the Information Gain (IG) value
• Determine if it was worthwhile to apply decision tree modeling to this dataset.
• IG is used to select the best attribute for splitting.

Confusion Matrix

• Complete the "Confusion Matrix" for Layer 2 based on the given decision tree model
• A confusion matrix summarizes the performance of a classification model.

Evaluation Metrics

• Calculate the following metrics using the completed "Confusion Matrix" for Layer 2:
  • Error Rate
  • Accuracy
  • Precision
  • Recall
  • F1 Score
• These metrics provide insights into the model's performance.

Variables for E-Commerce Company

• Identify the dependent variable (Label) for the e-commerce company, being whether a customer will make a purchase.
• Identify which other variables/attributes should be selected as independent variables.
• Independent variables are used to predict the dependent variable.