製造資料分析
===

[TOC]

Introduction
---

甚麼是異常檢測(Anomaly/Outlier detection)
---
:::warning
<style>
.blue {
color: blue;
}
.red {
color: red;
}
.orange {
color: orange;
}
.purple {
color: purple;
}
</style>
* 區分資料是<span class="blue">正常</span>還是<span class="red">異常</span>:tada::tada::tada:
:::

* <span class="red">異常檢測</span>的性質
* <span class="red">異常檢測</span>與正常資料有差異
* <span class="red">異常檢測</span>→<span class="purple">罕見資料</span>(==1%==)

* 那些情境會遇到<span class="red">異常資料</span>
* ==設備機台==、==網路瀏覽==、==信用卡紀錄==

基於統計/傳統機器學習所提出的異常檢測方法
---

### 理想狀況下的的異常檢測

>[Source from: Pattern Recognition “Anomaly Detection Challenges”
](https://www.researchgate.net/publication/321682378_Pattern_Recognition_Anomaly_Detection_Challenges)

### 基於統計/機器學習的異常檢測

:::success
* ==監督式==學習(supervised learning)
* 邏輯式回歸(logistic regression)
* 支持向量機(support vector machine)
* 隨機森林(random forest)
* XGBoost
:::danger
由於異常類別資料比例過少，通常我們需要作些調整! :fire:
:::spoiler

* 常見類別不平衡處理方法
* Up sampling
* Down sampling
* Weighted loss

:::
:::success
* ==非監督式==學習(unsupervised learning)
* 孤立森林(Isolation forest)
[Isolation forest-Novelty and Outlier Detection(Scikit-Learn)](https://scikit-learn.org/stable/modules/outlier_detection.html)

* Local Outlier Factors
* One-class SVM
:::

### 延伸閱讀 :mega:
:::info
* [機器學習 異常檢測 ( Anomaly detection )](https://kennyliblog.nctu.me/2019/08/30/Machine-learning-Anomaly-detection/#%E9%AB%98%E6%96%AF%E5%88%86%E5%B8%83-Gaussian-distribution)
* [異常檢測 — 從統計建模的角度切入吧](https://medium.com/@r41091113/%E7%95%B0%E5%B8%B8%E6%AA%A2%E6%B8%AC-%E5%BE%9E%E7%B5%B1%E8%A8%88%E5%BB%BA%E6%A8%A1%E7%9A%84%E8%A7%92%E5%BA%A6%E5%88%87%E5%85%A5%E5%90%A7-45e8ea9e2e25)
* [Data Mining常見的(異常檢測)算法有哪些？](https://www.zhihu.com/question/280696035)
:::

使用AutoEncoder作異常檢測
---

GAN與異常檢測的結合：GANomaly
---

[HackMD 使用教學](https://hackmd.io/c/tutorials-tw/%2Fs%2Fhow-to-embed-note-tw)

###### tags: `課堂筆記整理`