ISA 491 EXAM ONE TERMS & CONCEPTS

3/13/25

business analytics

process of transforming data into insights for informed decision-making

descriptive analytics

analyzing historical data to understand what has happened

• purpose: provide context and trends

• examples: sales reports, web traffic analysis

predictive analytics

using models to predict future outcomes

• purpose: anticipate trends and behaviors

• demand forecasting, churn prediction

prescriptive analytics

recommending actions based on predictive insights

• purpose: optimize decision-making

• examples: route optimization, personalized marketing

business intelligence

focuses on historical data and reporting

differences and similarities between BA and BI?

differences:

• BI is descriptive

• BA includes predictive modeling, forecasting and optimization

similarities:

• both leverage data for insights

data science

technical, focused on algorithms

differences and similarities between BA and data science?

differences:

• data science is more technical, focused on algorithms

• business analytics emphasized business context

similarities:

• both analyze data

data mining

discovering patterns in data

machine learning

algorithms that learn from data

artificial intelligence

broader concept of intelligent systems

big data

large, complex datasets requiring advanced tools

data-driven decision making

using data to guide actions and strategies

algorithm

A sequence of steps or rules followed to solve a problem or perform a computation. In machine learning, _________ process data to create models.

attribute, predictor, or input variable

Variables or features used to predict the outcome. These are independent variables in statistical models.

case, observation, record

a single unit of data in a dataset, often represented as a row in a table

categorical or factor variable

a variable that represents categories or groups

• examples: gender, product type, region

confidence

• statistical definition: A measure of certainty in an estimate, often represented by a confidence interval.

• machine learning definition: The probability assigned to a predicted class or outcome.

dependent, response, target, or outcome variable

the variable being predicted or explained in a model

• examples: sales revenue, customer churn, temperature

estimation

determining unknown parameters of a model based on data

prediction

using a model to forecast outcomes for new or unseen data

holdout data or set

a subset of data kept separate from training to evaluate a model’s performance on unseen data

inference

• statistical definition: a measure of certainty in an estimate, often represented by a confidence interval

• machine learning definition: the probability assigned to a predicted class or outcome

model

a mathematical representation of the relationships between variables

• in ML, models are built to make predictions or classification

conditional probability

the probability of an event occurring given that another event has occurred

• denoted as P(A|B)

prediction

the outcome or value that a model forecasts for given input data

profile

a description or summary of data for a single case or group of cases, often used for analysis or comparison

sample

a subset of data taken from a larger population, used for analysis

score

a numeric output from a model, often indicating the likelihood of a certain outcome

success class

the outcome of interest in a classification problem. For example, predicting “Yes” for a churn model

supervised learning

a type of machine learning where the model is trained on labeled data (i.e., inputs paired with known outputs)

test data or set

data used to evaluate the final performance of a model after training and validation

training data or set

data used to build and train a model, including identifying relationships and adjusting parameters

unsupervised learning

a type of machine learning where the model learns patterns and structures from unlabeled data

validation data or set

data used during training to tune model parameters and avoid overfitting

variable

a measurable characteristic or attribute in a dataset.

- can be dependent, independent, categorical, or continuous

R

A programming language for statistical computing and graphics.

• Open-source and widely used in data analysis and machine learning.

• Command-line interface for running code.

RStudio

An integrated development environment (IDE) for R.

• Provides a user-friendly interface with features like:

  • Script editor

  • Console

  • Environment pane

  • Plot viewer

package (in R)

a collection of functions, datasets, and compiled code that extends the functionality of R

literate programming

a programming paradigm introduced by Donald Knuth that combines human-readable text with executable code.

core principle of literate programming

“Code is written for humans to read and only incidentally for machines to execute.”

core ideas of machine learning

1. classification

2. prediction

3. association rules & recommenders

4. data & dimension reduction

5. data exploration

6. visualization

7. generative AI

classification

Predicting categorical target variables using algorithms like decision trees and logistic regression

prediction

Estimating numerical values using methods like linear regression and neural networks

association rules & recommenders

Identifying relationships between variables to recommend items based on purchasing behavior (e.g., Apriori algorithm)

data & dimension reduction

Simplifying datasets into more manageable forms by reducing or combining variables with techniques like Principal Component Analysis (PCA)

data exploration

summarizing and visualizing data to uncover patterns, identify anomalies, and refine questions for analysis

visualization

conveys information and insights effectively using charts, graphs, and plots

generative AI

involves creating new data instances that resemble the input data

steps for machine learning

1. understand purpose of project

2. obtain data

3. sample (optional)

4. explore, preprocess, and prepare data

5. reduce dimensions

6. partition data (for supervised tasks)

7. choose machine learning techniques and apply them

8. interpret and assess results

9. deploy the model

clustering

common unsupervised learning technique used to group similar data points into clusters based on their feastures

overfitting

occurs when a model learns not only the underlying patterns in the data but also the random noise or idiosyncrasies

cross-validation

statistical technique used to evaluate a model’s ability to generalize to new data

• take training data and divide into certain # of folds

• calculate holdout R² / RMSE for each fold

• average values

Root Mean Squared Error

Represents the square root of the average squared differences between predicted and actual values

• A lower value indicates better model performance

R-squared

Indicates the proportion of variance in the target variable that is explained by the model

• Higher values (closer to 1) suggest better explanatory power of the model

purposes of exploratory data analysis

1. Exploration and Data Preparation: Investigating the dataset to identify patterns, anomalies, and necessary preprocessing steps for modeling.

2. Presentation and Storytelling: Using visualizations and summaries to communicate insights, often for stakeholders or reports.

variable encoding in data analysis

representing the data in a specific format

simple imputation

replace the missing values with a constant value based on the data

• examples: mean, median, mode, constant

time series imputation

replace using the time order of the data

interpolation

estimate missing values using interpolation

predictive imputation

use a model to predict the missing value based on other variables in the data set

• examples: linear regression, k-nearest neighbors (KNN)

advanced methods of imputation

• examples: multiple imputation, maximum likelihood estimation

outlier

data point that is significantly different from other observations in the dataset

dimension (p) of a data set

number of variables (columns) of the data set

principal components analysis (PCA)

method that uses geometry to create a new coordinate system for the data based on the correlation structure

• first coordinate contains the most variation (information) in the data; second contains the second most variation, etc.

total variation of PCA

sum of the variances of each variable

covariance

measure of the unscaled linear association between two variables

evaluating explanatory models

1. Theoretical Justification: Does the model align with existing theories and concepts?

2. Model Fit: How well does the model fit the training sample?

3. Variable Significance: Which predictors are statistically significant?

4. Model Interpretability: Can the model results be easily understood and communicated?

5. Hypothesis Testing: Can the model be used to test specific hypotheses or theories?

evaluating predictive models

1. Prediction Accuracy: How well does the model predict new observations (in validation/test samples)?

2. Generalization: How well does the model perform on unseen data?

3. Model Complexity: Is the model simple enough to avoid overfitting?

4. Business Value: Does the model provide actionable insights and support decision-making?

three uses of predictive models

1. prediction: estimates a continuous numerical value based on input data

2. classification: assigns a categorical label to an observation based on input features

3. propensity or ranking: predicts the likelihood or ranking of outcomes rather than a direct value or label

predictive accuracy

measures how well the model predicts new observations

• essential for assessing the model’s generalization to unseen data

naive benchmark

simplest form of predictive performance evaluation

mean absolute error (MAE)

measures the average magnitude of prediction errors without considering their direction

mean error

captures the average error, retaining the signs of the errors(positive or negative)

mean percentage error (MPE)

gives the percentage score of how predictions deviate from the actuals (on average)

mean absolute percentage error (MAPE)

measures the average percentage deviation of predictions from actual values

root mean squared error

similar to the standard error estimate in regression, but is computed on the holdout sample rather than the training sample

training data & errors

data: used to train the model
errors: called residuals in regression

holdout data & errors

data: reserved for evaluation to mimic real-world performance
errors: computed by comparing the predicted to actual values on new data

cumulative gains curve

shows the proportion of the actual number of positive cases (y-axis) that are captured by considering a certain proportion of the dataset (x-axis)

• useful for evaluating the model’s ability to rank cases in order of performance

lift

measures the ratio of the cumulative gains of the model to the cumulative gains of a baseline model

• shows how much better the model is at identifying high-value cases compared to random selection

oversampling

method to over represent the class of interest; will improve ability to develop a model that predicts our class of interest

using MLR for prediction vs explanation

predictive model:

• how well the model will perform on new data

• measures the fit between the model and new, unseen data

• focuses on small prediction errors (we want to minimize the difference between the predicted and actual values of response variable)

explanatory model:

• good model fits the data well and has interpretable coefficients

• uses the entire dataset to estimate the best-fit model

• measures the fit between the model and the training data

forward selection

adds variables sequentially to the model based on the next best variable

backward selection

removes variables sequentially from the model based on the next worst variable

stepwise selection

combines forward and backward selection; variables are added and possibly removed at each step

regularization or shrinkage

method where we shrink the coefficients toward zero; imposes a penalty on the model fit

two methods of regularization

Lasso and Ridge

ridge regression

the penalty is based on the sum of the squared regression coefficients, ∑𝑝𝑗=1𝛽2𝑗∑j=1pβj2. This is called L2 regularization

lasso regression

the penalty is based on the sum of the absolute values of the regression coefficients, ∑𝑝𝑗=1|𝛽𝑗|∑j=1p|βj|. This is called L1 regularization

ordinary least squares regression (OLS)

we select the estimated coefficients by minimizing the training SSE:

∑𝑖=1𝑛(𝑦𝑖−𝑦̂ 𝑖)^2

elastic net

an approach that combines L1 and L2 and has two tuning parameters, α and λ

alpha (α)

a mixing parameter that determines the mix of L1 and L2 regularization

• equal to 0: elastic net is equivalent to ridge regression

• equal to 1: elastic net is equivalent to lasso regression

• between 0 and 1: elastic net is a combination of ridge and lasso regression

λ

tuning parameter that controls the overall amount of shrinkage