Please enable JavaScript.
Coggle requires JavaScript to display documents.
Session 6 - Introducing Machine Learning - Coggle Diagram
Session 6 - Introducing Machine Learning
Data-driven decision making
Data mining
= process of extracting knowledge from data;
find patterns, trends, and anomalies in data
Response to "data-explosion"
(socio-demographic information, Business transactions...)
-> increase of volume, variety & veracity and so increase of the need of tool/techniques to extract meaningful insights
Purpose
= extract interesting, non-trivial, implicit, previously unknown, & potentially useful patterns or knowledge from huge amounts of data
KDD
= transforming data into knowledge
4 steps process
Selection
: 1st step = select the relevant data for the analysis (cleaning & pre-processing the data, error removed)
Preprocessing
: 2nd step is to prepare the data for mining (transforming the data into a diff format)
Data mining
: 3rd step = apply data mining algorithms to the data to extract patterns & knowledge
Interpretation/Evaluation
: Last step = interpret & evaluate the results/significance of the data mining process
Descriptive & Predictive Analytics
Descriptive
= describe what has happened in the past
uses
summary statistics
-> average, counts, median, variance, relative frequency, counts by a certain group, etc.
Ex
: T nbr of cust. last year; % of cust. that spend > 100€; % of female cust.; average spending male customers
Type of data
: Structured data
(cust. records, sales transactions, website logs)
& semi-structured data
(XML, JSON, email messages, web pages)
, such as transactional data, customer data, and website data.
Predictive
= predict what will happen in the future w/ available data
Ex
: Predicting cust. churn; Predicting conversions (=predicting whether a customer will convert based on its characteristics)
Type of data
: Structured & unstructured data (
emails, social media posts, customer reviews, medical images)
Machine Learning
= application of AI allowing computers to learn automatically without human intervention
Supervised
= training a machine from labeled data
Machines learn the relationship between inputs (fruit images) & outputs (fruit labels
Trained machines can make predictions on new, unlabeled data
Classification
: dependent var. is class/binary
Estimation
(or
Regression
): dependent var. is continuous
Unsupervised
= training a machine from unlabeled data
Artificial Intelligence
Narrow AI
= focused on a specific task
(credit scoring, customer retention scoring, response models)
General AI
= capable of performing any intellectual task as it has the ability to learn & adapt
Super AI
= out performs human intelligence
Classification
= identifying the category or class of a data point
Applications
Customer Attrition or Churn
: Ex:
Goal
= Predict whether a customer is likely to churn or leave for a competitor;
Approach
= Label customers as loyal or disloyal
Fraud Detection
:Ex:
Goal
= predict fraudulent cases in credit card transactions;
Approach
= use credit card transaction data and account holder information as attributes
Digital Marketing
: Ex:
Goal
= Reduce the cost of direct mailing by targeting a set of consumers likely to purchase a new cellphone;
Approach
= Use historical data for a similar cellphone product to identify customers who purchased and those who did not
= Training set w/ data -> Learn classifier -> Modeling <- Test classifier w/ table of training data now classified
Regression
= predicting a continuous numerical outcome
Ex: Predicting sales amounts of new product based on advertising expenditure; Predicting wind velocities as a function of temperature, humidity, air pressure, etc.; Time series prediction of stock market indices
Clustering
= grouping data points together based on their similarity
Market Segmentation
= process of dividing a market into groups of customers with similar characteristics
can be done using clustering algorithms to group customers together based on their demographics, lifestyle, and purchase history.
Document Clustering
= grouping documents together based on their content
can be done using clustering algorithms to group documents together based on their keywords, topics, and sentiment.
Algorithms
Linear Regression
Linear regression is a supervised learning algorithm that is used to predict a continuous numerical outcome. It is a simple and effective algorithm that is often used for tasks such as predicting house prices, sales, or risk of churn.
Logistic Regression
Logistic regression is a supervised learning algorithm that is used to predict a categorical outcome. It is a more complex algorithm than linear regression, but it can be more accurate for predicting categorical outcomes. Logistic regression is often used for tasks such as spam filtering, medical diagnosis, or customer segmentation.
Decision Trees
Decision trees are a type of supervised learning algorithm that is used to make predictions by splitting the data into smaller and smaller groups. Decision trees are often used for tasks such as classification and regression. They are a good choice for problems where there are no clear linear patterns in the data and interaction effects between the variables.
Deep Learning Models
Deep learning models are a type of machine learning algorithm that is inspired by the structure of the human brain. They are composed of multiple layers of artificial neurons that are interconnected. Deep learning models are able to learn complex patterns from data that would be difficult or impossible to learn with traditional machine learning algorithms.
Deep Learning Models for Generative AI
Deep learning models are often used to train large language models (LLMs) that can generate human-quality text. LLMs are the basis for many generative AI applications, such as chatbots, machine translation, and text summarization.
Feature Engineering
Feature engineering is the process of transforming raw data into a format that is suitable for machine learning algorithms. This can involve tasks such as data cleaning, feature extraction, and feature scaling. Feature engineering is an important step in the machine learning process, as it can significantly impact the performance of the model.
Predictive Modeling Timeline
= Use historical customer behavior data to predict future customer behavior
Independent period
(historical customer behavior)
= when the customer is not dependent on the company
Dependent period
(ex: did cust. churn?)
= when the customer is dependent on the company
can be used to improve customer segmentation, target marketing campaigns, & reduce customer churn