Business Analytics: Concepts and Methods (Chapter 14 Study Notes)

Business Analytics: Concepts and Methods

Introduction to Business Analytics

Business analytics refers to the systematic analysis of data to inform business decision-making. It encompasses a range of statistical and computational methods to extract insights, predict future trends, and optimize business processes.

• Business analytics is divided into three main categories: descriptive, predictive, and prescriptive analytics.

• Each category addresses different types of business questions and utilizes distinct statistical methods.

Major Categories of Business Analytics

• Descriptive Analytics: Answers "What has happened or has been happening?" by summarizing historical data to identify patterns and trends.

• Predictive Analytics: Answers "What could happen?" by using historical data and statistical models to forecast future events.

• Prescriptive Analytics: Answers "What should happen?" by recommending actions to optimize outcomes and manage future circumstances.

Descriptive Analytics

Descriptive analytics focuses on summarizing and visualizing historical data to provide insights into business activities.

• Key Point: Enables decision makers to "drill down" into data details for deeper understanding.

• Example: Dashboards provide real-time summary displays for monitoring business activities.

• Data Dimensionality: Descriptive analytics often involves high-dimensional data, visualized using color, size, or motion to represent multiple variables.

• Visualization Techniques: Scatter plots and dynamic bubble charts can display additional dimensions of data.

Predictive Analytics

Predictive analytics uses statistical models and historical data to forecast future events and trends.

• Key Point: Relies on inferential statistics, such as regression analysis, to make predictions.

• Methods:

  • Prediction: Assigns a value to a target variable based on a model.

  • Classification: Assigns items to target categories or classes.

  • Clustering: Finds natural groupings in data.

  • Association: Identifies items that tend to co-occur and specifies rules for their co-occurrence.

• Example: Regression trees and classification trees are used to split data into groups for prediction and classification tasks.

Supervised and Unsupervised Methods

Business analytics methods can be classified as supervised or unsupervised, depending on the use of training data and explicit goals.

• Supervised Methods:

  • Begin with explicit facts (training data) to build models.

  • All inferential methods discussed are supervised.

  • Overfitting: Occurs when a model describes random error rather than underlying relationships; techniques like cross-validation help prevent this.

• Unsupervised Methods:

  • Build models without training data or explicit goals.

  • Variable selection is critical to avoid overfitting.

  • Models may or may not be useful for business decision-making.

  • Frameworks like DCOVA (Define, Collect, Organize, Visualize, Analyze) improve model usefulness.

Regression and Classification Trees

Regression and classification trees are predictive analytics methods that split data into groups based on independent variable values.

• Regression Trees: Used for predicting continuous dependent variables.

• Classification Trees: Used for predicting categorical dependent variables.

• Splitting Criteria:

  • Akaike Information Criterion (AIC): $AIC = -2 \ln(L) + 2m$

  • Corrected AIC (AICc): $AIC_c = AIC + \frac{2m(m+1)}{n-m-1}$

  • LogWorth Statistic: $\text{LogWorth} = -\log_{10}(p)$, where $p$ is the adjusted p-value for the split.

  • Smaller AIC/AICc values indicate better models; LogWorth > 2 suggests a split should be made.

• Example: A regression tree may split data first by Price < 0.99, then by Price < 0.79, reporting mean sales for each subset.

Clustering Methods

Clustering groups items based on similarity, often measured by Euclidean distance.

• Euclidean Distance Formula: $d_{ij} = \sqrt{\sum_{k=1}^{r} (x_{ik} - x_{jk})^2}$

  • $d_{ij}$ = distance between object $i$ and object $j$

  • $x_{ik}$ = value of object $i$ in dimension $k$

  • $x_{jk}$ = value of object $j$ in dimension $k$

  • $r$ = number of data dimensions

• Example: Cluster analysis may reveal that one cluster has higher means for Sharpe Ratio and returns, and lower Expense Ratio.

Association Methods

Association methods uncover patterns among items, often using specialized techniques for categorical or numerical variables.

• Multiple Correspondence Analysis (MCA): Best for categorical variables; uses biplots to visualize relationships in contingency tables.

• Multidimensional Scaling (MDS): Best for numerical variables; visualizes associations in two or more dimensions using distance measures.

• Example: MDS can show that certain sports are perceived as more similar to each other based on survey data.

Text Analytics

Text analytics deals with unstructured text data, blending descriptive and prescriptive analytics to automate interpretation.

• Unstructured text: Includes words, phrases, and passages that do not fit a template.

• Latent Semantic Analysis: Clusters text based on latent dimensions of similarity.

• Example: Analyzing social media comments to identify emerging topics or sentiment.

Prescriptive Analytics

Prescriptive analytics recommends actions to optimize business performance and manage future circumstances.

• Combines: Traditional statistical methods, management science, and information systems.

• Methods: Optimization and simulation are commonly used.

• Example: Using simulation to test different business strategies and recommend the best course of action.

Summary Table: Predictive Analytics Methods and Their Uses

Method	Prediction	Classification	Clustering	Association
Classification & Regression Trees	✓	✓
Cluster Analysis			✓
Multidimensional Scaling (MDS)				✓

Chapter Summary

• Business analytics is essential for modern decision-making, encompassing descriptive, predictive, and prescriptive methods.

• Understanding inferential statistics is crucial for effective predictive analytics.

• Supervised and unsupervised methods offer different approaches to model building and data analysis.

• Visualization and dimensionality reduction techniques enhance the interpretability of complex data.