Predictive Analytics

Course code : 21AI743 (VTU 2021)

Instructor: Achyuth Kasyap P

[email protected]
Introduction
Predictive analytics involves a
range of statistical methods,
including modeling, machine
learning, and data mining,
which are used to analyze
current and historical data to
forecast future outcomes. Image credit: [1]

Business analytics involves the use of

skills, technologies, and practices for the
ongoing exploration and analysis of past
business performance, aimed at gaining
insights and supporting business planning.
Types of Business Analytics
Descriptive Analytics Techniques such as data visualization,
along with statistical measures like
Descriptive analysis focuses mean, median, and mode, are commonly
on summarizing collected data used to perform descriptive analytics.
to provide a clear and Image credit: [2]
straightforward overview,
explaining key observations
and patterns within the data.
Predictive Analytics An example of predictive analytics is a
This involves a range of retailer forecasting future product
statistical techniques, including demand during the holiday season based
modeling, machine learning, on past sales data.
and data mining, used to
analyze current and historical
data in order to predict future Image credit: [3]
outcomes.
Prescriptive Analytics An example of prescriptive
Prescriptive analytics extends beyond analytics is a logistics
predicting outcomes by providing company using it to identify
recommendations or strategies based the most efficient delivery
on data analysis. It guides routes, minimizing both
decision-making by suggesting the costs and delivery time.
best course of action to achieve
Image credit: [4]
specific goals.
Applications of Business Analytics
Clinical decision support Analytical CRM
systems A commercial application of predictive
Predictive analytics in healthcare is analytics involves analyzing and
primarily used to identify patients at understanding products that are in
risk of developing conditions such as demand or have the potential for high
diabetes or other chronic diseases, demand.
enabling early intervention and
better management.

Image credit: [5]

Collection Analytics Direct Marketing
This analysis helps identify Predictive analytics can determine
customers who miss payment the optimal mix of product
deadlines, allowing for the versions, marketing materials,
efficient allocation of resources communication channels, and
and determining the most timing to best target a specific
effective legal methods for consumer.
collecting payments. Image credit: [6]
Analytical techniques
Regression Based Machine learning
These models aim to create a In this case, an equation is also used,
mathematical equation that but it can be a nonlinear function
represents the interactions that is continuously updated
between various variables. throughout the training process.
Regression based techniques
Linear Regression Logistic Regression
This model examines the relationship In this method, we predict a probability,
between a dependent variable and a which serves as the dependent variable, by
group of independent or predictor utilizing the independent variables or
variables, aiming to understand how features. This prediction is made through
changes in the predictors influence the either a linear or nonlinear function,
outcome. By analyzing these depending on the complexity of the data and
relationships, the model can provide the relationships between variables. The
insights into patterns and make informed approach allows for modeling the likelihood
predictions. of specific outcomes based on input factors.
Regression based techniques
Time Series Models Classification and Regression Trees
These models are designed to This technique generates either
predict or forecast the future classification trees or regression trees,
behavior of variables by analyzing depending on the nature of the dependent
data points collected over a period of variable. If the dependent variable is
time. By accounting for trends, categorical, it produces a classification tree,
patterns, and fluctuations within the whereas if the dependent variable is
data, they provide valuable insights numeric, it results in a regression tree. This
into how variables are likely to method helps in identifying patterns and
evolve in the future. relationships within the data to make
informed decisions or predictions.
Machine learning techniques
Neural Networks Support Vector Machines
Neural networks are advanced, nonlinear These techniques are used to uncover
modeling techniques capable of capturing and utilize complex patterns in data by
and modeling complex functions. They clustering similar data points and
consist of interconnected nodes or "neurons" ranking them according to specific
that process data through multiple layers, attributes. This method allows for the
allowing them to learn intricate patterns and discovery of underlying structures and
relationships within the data. This relationships, leading to more effective
sophistication enables them to tackle a wide analysis and strategic insights.
range of challenging problems and tasks.
Machine learning techniques
Naive Bayes K-Nearest Neighbors
This method assumes that the predictors are In this approach, no formal training is
statistically independent, which simplifies required. Instead, a sample is classified by
interpretation. It is grounded in the concept of measuring its distance to the nearest
probability, leveraging this assumption to make neighbors in the dataset. This method
predictions or classify data based on the relies on the principle that similar samples
likelihood of various outcomes. This approach are likely to be close to each other in the
provides a clear and straightforward way to feature space, and thus, classification is
understand and analyze the relationships based on the proximity to other known
between variables. samples.
Tools for Predictive Analytics
There are numerous tools available in the marketplace that help with the execution of predictive
analytics. These range from those that need very little user sophistication to those that are
designed for the expert practitioner.

● Scikit-learn
● KNIME Image credit: [7]
● Open NN
● R
● Python
● Keras
● Pytorch
● Pandas
● Matplotlib
Predictive Modeling
Predictive modeling is a technique in both statistics
and machine learning that forecasts future outcomes
by analyzing historical data. This approach can be
applied in various scenarios, such as predicting sales
trends, forecasting customer behavior, estimating the
likelihood of equipment failures, or evaluating Image credit: [8]
financial risks. By leveraging past data, predictive
modeling helps anticipate future events and make
informed decisions.
Propensity models are used in data Unlike classification clustering models
science and marketing analytics to segment data into groups that were not
estimate the likelihood that an individual previously defined with the help of
or group will take a specific action. various mathematical algorithms.
Propensity Models
Propensity Customer Lifetime Value Propensity to Engage
It is an estimate of the total value a business A propensity-to-engage model predicts the
will gain from its entire relationship with a likelihood that a person will engage in
customer. This prediction forecasts the total specific activities, such as unusual behavior
amount a customer is expected to spend with a following a purchase. This model helps
company over their lifetime, providing anticipate which individuals are more likely
insights into long-term revenue potential and to exhibit certain behaviors, allowing
guiding strategies for customer retention and businesses to tailor their strategies and
growth. interventions accordingly.
Propensity Models Propensity to Churn
The propensity-to-churn model identifies active
customers who are at risk of leaving, allowing you
Propensity to Buy to focus on high-value customers who can
The propensity-to-buy model potentially leave.
identifies which customers are most
likely to make a purchase, enabling
you to determine which customers to Behavioral Clustering
target effectively. By understanding This clustering analysis provides insights into
which individuals are on the verge of purchasing behavior by examining factors such as the
buying, businesses can focus their frequency of purchases, the amount spent, and the time
marketing efforts and resources more interval between subsequent purchases. By
efficiently. understanding these patterns, businesses can better
predict customer behavior.
Cluster Models
Product based Clustering Brand based Clustering
This clustering helps uncover which This approach focuses on the brands of
product groups customers are more likely items customers purchase. Marketers can
to purchase, enabling more effective leverage this information to predict which
product planning and strategy. By other brands customers are likely to buy,
understanding these preferences, allowing for more targeted and effective
businesses can optimize their product marketing strategies. By understanding
offerings and make informed decisions to brand preferences, businesses can tailor
meet customer demand and improve their campaigns and promotions to better
overall efficiency. align with customer interests.
Applications of Predictive Analytics
Uplift Modeling Auto Insurance
It is a technique for modeling the change in Predictive modeling is utilized in vehicle
probability caused by an action. For example insurance to assign risk of incidents to from
in a retention campaign you wish to predict the information obtained by policyholders.
change in probability that customer will
remain a customer if they are contacted.

HealthCare Customer relationship management

Here predictive modeling is used to predict Predictive modeling is used extensively in
and analyse the usage and claims of health analytical customer relationship management
insurance based on customer’s data and issues. and data mining to produce customer-level
models.
References
[1] https://www.clarkson.edu/academics/majors-minors/mba-business-analytics

[2] https://www.investopedia.com/terms/d/descriptive-analytics.asp

[3] https://www.investopedia.com/terms/p/predictive-analytics.asp

[4] https://www.linkedin.com/pulse/descriptive-predictive-prescriptive-analytics-2023-marc-mandel-ccxp

[5] https://news.twentyci.co.uk/blog/kick-start-a-crm-programme-with-three-simple-steps

[6] https://www.sender.net/blog/direct-marketing/

[7] https://www.educba.com/predictive-analytics-tools/

[8] https://www.educba.com/predictive-modeling/