Man1130/jupyter/Man1130-python-comission/slides/slides.md

marp, title, description, theme, paginate, _paginate, backgroundImage, footer, style

marp	title	description	theme	paginate	_paginate	backgroundImage	footer	style
true	Marp CLI example	Hosting Marp slide deck on the web	uncover	true	false	url('https://www.google.com/url?sa=i&url=https%3A%2F%2Fcommons.wikimedia.org%2Fwiki%2FFile%3AHelloWorld.svg&psig=AOvVaw0d3lmyaMphPi0ANeGIEJOw&ust=1670049479380000&source=images&cd=vfe&ved=0CBAQjRxqFwoTCJjxx6Gp2vsCFQAAAAAdAAAAABAE')	2022 project presentation	section { background-color: #ccc; padding: 0 10vw; } footer { text-align: center; }

slide topic?

overview of the dataset

• Male and Female proportion
  • male by age
  • female by age
• The types of chest pain experienced among the patients
  • pie chart
• showing the correlation between chest pain and target Confusion Matrix ?

---

Decision Tree

overview of the dataset

• Decision Tree 果張圖
  • Performance Analysis
    • confusion matrix
    • ROC

---

Naive Bayes

overview of the dataset

• Naive Bayes 果張圖
  • Performance Analysis
    • confusion matrix
    • ROC

---

Logistic Regression

overview of the dataset

• Logistic Regression 果張圖

---

Performance Analysis

• Performance Analysis
  • confusion matrix
  • ROC

---

Performance Analysis con't

• pick a sample, bayes modeling:

  • naive_bayes_scramble.ipynb
  • column selection
    • compare column vs accurancy

• the performance/accurancy of fewer column MAY BE better than more column

  • possible cause
    • column noise/ input data accurancy ?
    • modal overfitting ?
    • extreme case ?

---

Performance Analysis con't

• how to improve ?
  • the choice of the columns may be better if other facuity involved.
  • more labelled data improves accuracy

---

Performance Analysis con't

• disclaimer ?

  • no model can introduce 100% accurancy
  • why ?
    • extreme case
    • chaos theory ?
      • will never take all ervery matters into account

• however, the model can be considered if accuracy above nn% in general

---

why these method ?

• unsupervised modeling

  • cross out reason

    • data is already labelled
    • data is small amount and discrete

  • Knn

  • Kmeans

---

why these method ?

• supervised modeling / supervised learning

  • data is already labelled

    • []Decision Tree
    • []Naive Bayes
      • performance ? accuracy ? ROI ?

  • multi dimensional/column difficult to understand/maintain

    • Logistic Regression
    • SVM

---

notes only