0% found this document useful (0 votes)
6 views26 pages

Supervised Learning Techniques Explained

Uploaded by

adambassoumi5
Copyright
© All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
6 views26 pages

Supervised Learning Techniques Explained

Uploaded by

adambassoumi5
Copyright
© All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd

THAT’S IT FOR TODAY!

Supervised Learning
Classification
Regression vs Classification
Business applications
Models: Logistic Regression, Naive Bayes, k-NNs
Terminology: from variable to early stopping
Mathemical explanation for each of core models
Model choice based on on business questions and data
Initiation - evaluation - fitting - reprocessing - fine-tuning
Python code for each stage of the process

Julia Lenc
Analytics Journey
Business Analytics (BA)
1. Intro: Business and Revenue models. KPIs
2. Business models translated into analytics
3. Techniques: Descriptive, Diagnostic, Predictive, Prescriptive

Diagnostic Techniques
1. Inference: hypotheses testing
2. Unsupervised Learning: clustering, dimensionality reduction, anomalies

Predictive Techniques
1. Supervised learning: overview
2. Preparation: data pre-processing
3. Foundations: model choice and evaluation
4. Regression: linear and non-linear
5. Classification: logistic regression, Naive Bayes, k-NNs (this presentation!)
6. Time series: ARIMA, SARIMA, Exponential Smoothing
7. Advanced: Decision Trees, SVM
8. Ensemble: bagging, boosting, stacking
9. Neural Networks: FFNN, CNN, RNN, Transformers

Prescriptive Techniques
1. Optimization: Linear, Non-linear and Dynamic programming
2. Simulation: Monte Carlo, Discreet Events, System Dynamics
3. Probabilistic Sequence: Markov Chains, Markov Decision Processes
4. Reinforcement Learning: Q-Learning, Deep RL, Policy Gradient

Julia Lenc
THAT’S IT FOR TODAY!

The process: Stage 1


1. Understand and formulate the business question as:

Specific: Define the problem clearly (e.g., classification vs. regression tasks).

Measurable: Determine evaluation criteria (e.g., accuracy, precision / recall).

Achievable: Ensure data and resources match the modeling goals.

Relevant: Validate that answering this question will drive action.

Time-bound: Identify stakeholders and key deadlines for delivery.


A cost function used to evaluate probabilistic models, penalizing incorrect predictions heavily.

2. Prepare the data to meet requirements:

Read here about 8 stages of data preparation to get perfect input

3. Choose the model and set evaluation criteria:

Read here about confusion matrix, evaluation criteria and “precision vs recall”

Julia Lenc
THAT’S IT FOR TODAY!

Import the data


Pre-requisites: data preprocessing is here, validation criteria are here

A cost function used to evaluate probabilistic models, penalizing incorrect predictions heavily.
THAT’S IT FOR TODAY!

Model Run. Evaluation

A cost function used to evaluate probabilistic models, penalizing incorrect predictions heavily.

Evaluation

Rest of the code is on the next slide :)


THAT’S IT FOR TODAY!

Model Run. Evaluation

A cost function used to evaluate probabilistic models, penalizing incorrect predictions heavily.

This is the rest of the code from previous page


Evaluation

It would not work on its own :)


THAT’S IT FOR TODAY!

Model Run. Evaluation

A cost function used to evaluate probabilistic models, penalizing incorrect predictions heavily.

This is the rest of the code from previous page


Evaluation

It would not work on its own :)


Here we evaluate accuracy. Add other aligned metrics!
THAT’S IT FOR TODAY!

Overfitting / Underfitting

A cost function used to evaluate probabilistic models, penalizing incorrect predictions heavily.

Evaluation
THAT’S IT FOR TODAY!

Reassess pre-processing

A cost function used to evaluate probabilistic models, penalizing incorrect predictions heavily.

Evaluation
THAT’S IT FOR TODAY!

Fine-tuning
(last chance before complexity increase)

A cost function used to evaluate probabilistic models, penalizing incorrect predictions heavily.

Evaluation

You might also like