Business Intelligence (BI) And Business Analytics Flashcards
The ability to gather and make sense of information in the context of a business
Business intelligence
Purpose of business intelligence
Gain superior insight and understanding of the business and it’s ecosystem
Understand the past and the present -> predict the future
Make better decisions
Components of business intelligence
Storage (data warehouse/data marts)
Data mining tools (for business analytics)
Reporting and visualization tools (e.g. Dashboards)
What is data mining?
The computational process of discovering patterns in large data sets involving methods at the intersection of artificial intelligence, machine learning, statistics, and database systems
Customer segmentation
Who are the most valuable customers to a girl
Marketing and promotion targeting
Identifying which customers will respond to each offer
Market basket analysis
Which products customers buy together and how an organization can use this information to cross sell more
Type of association rules mining determining what products go together in a shopping cart at a retailer
Collaborative filtering
Personalizing an individual customers experience based on trends and preferences exhibited by similar customers
Customer churn
Which customers are more likely to leave and which retention strategies are most likely to succeed
Fraud detection
Uncover patterns consistent with criminal activity
Financial modeling
Building trading systems that adapt to historical trends or risk models to identify customers with the highest likelihood to default on a credit
Five classes of data mining tasks
Association detection (can be both) Clustering (unsupervised) Classifications (supervised) Regressions (supervised) Anomaly/outlier
Unsupervised data mining
Analysts do NOT create the model before running analysis
Apply data mining technique and observe results
Hypothesis created AFTER analysis as explanation for results
Ex. Cluster analysis, cluster creation for collaborative filtering
Supervised data mining
Model developed BEFORE analysis
Statistical techniques used to estimate parameters
Ex. Classification, regression analysis
Association rules mining
Determine which behaviors/outcomes go together
Find relationships among attributes in data that frequently occur together
Ex. Products bought together, symptoms and illnesses manifest together
Product affinities
Likelihood of two or more products being sold together
Support (association rule evaluation)
How often do these things appear together?
The probability that things will occur together
s{product1, product2}
SLHS&RHS
Confidence (association rule evaluation)
Given LHS, how often do we see RHS?
P(RHS|LHS)
sLHS&RHS/sLHS
*asymmetric
Lift (association rule evaluation)
How often does LHS appear with RHS, compared to how often chance would predict RHS would occur anyway?
The ratio of observed support to the expected support assuming the events are independent
c(LHS->RHS)/s(RHS)
(sLHS&RHS/sLHS)/sRHS
>1 indicates positive correlation (co occurance more likely than chance)
= approx 1 indicates almost no correlation, events are independent
<1 indicates negative correlation - co occurrence is less likely than chance
Complementary products lift
Greater than 1
Substitute products lift
Less than 1
Cluster analysis
Similar records (or characteristics) are grouped together Does not rely on predefined categories (labels, groups) - records grouped together on the basis of self-similarity (unsupervised data mining)
Classification
Arrange the data into predefined groups (supervised data mining)
Recursive partitioning
A technique for creating a decision free to reach the desired level of purity
Purity of a subgroup
The proportion of its records that belong to the same class
Error rate
Percent of misclassified records out of the total records in the validation data
Positive Predicted, Positive Actual
True positive
Positive predicted, negative actual
False positive
Negative predicted, positive actual
False negative
Negative predicted, negative actual
True negative
N (sum of predictions)
TP + FP + FN + TN
Accuracy
How often is the classifier correct?
(TP+TN)/N
1 - error rate
Misclassification rate (how often is it incorrect?)
1 - accuracy
Precision
When it predicts positive, how often is it correct?
TP/(TP+FP)
Specificity
When it predicts negative, how often is it correct?
TN/(TN+FN)
When might a model with greater total error be chosen?
The cost of one kind of misclassification is unacceptably high
Credit default, computer network intrusion, national security risk, presence of cancer
