Upgrade to Pro
— share decks privately, control downloads, hide ads and more …
Speaker Deck
Features
Speaker Deck
PRO
Sign in
Sign up for free
Search
Search
Confusion matrix
Search
Sunmi Yoon
November 03, 2019
Technology
170
0
Share
Confusion matrix
Confusion matrix 기초부터 머신러닝 응용까지 for dataitgirls3
Sunmi Yoon
November 03, 2019
More Decks by Sunmi Yoon
See All by Sunmi Yoon
데이터 분석가 채용 공고 읽는 방법
ysunmi0427
1
380
Deep down in classification 0.5 magic number
ysunmi0427
0
110
Tree Methods
ysunmi0427
0
130
심슨의 역설
ysunmi0427
0
2.5k
회사는 어떤 사람을 데이터 분석가로 채용하고 싶어하는 것일까?
ysunmi0427
0
2.5k
Other Decks in Technology
See All in Technology
20260410 - CNTUG meetup #72 - DiskImage Builder 介紹:以 Kubespray CI 打造 RockyLinux 10 Cloud Image 為例
tico88612
0
120
ログ基盤・プラグイン・ダッシュボード、全部整えた。でも最後は人だった。
makikub
5
1.7k
3つのボトルネックを解消し、リリースエンジニアリングを再定義した話
nealle
0
400
Sansan Engineering Unit 紹介資料
sansan33
PRO
1
4.2k
ある製造業の会社全体のAI化に1エンジニアが挑んだ話
kitami
2
910
#jawsugyokohama 100 LT11, "My AWS Journey 2011-2026 - kwntravel"
shinichirokawano
0
130
Introduction to Sansan, inc / Sansan Global Development Center, Inc.
sansan33
PRO
0
3k
TanStack Start エコシステムの現在地 / TanStack Start Ecosystem 2026
iktakahiro
1
380
2026年度新卒技術研修 サイバーエージェントのデータベース 活用事例とパフォーマンス調査入門
cyberagentdevelopers
PRO
7
7.9k
システムは「動く」だけでは足りない 実装編 - 非機能要件・分散システム・トレードオフをコードで見る
nwiizo
3
350
非エンジニア職からZOZOへ 〜登壇がキャリアに与えた影響〜
penpeen
0
370
名刺メーカーDevグループ 紹介資料
sansan33
PRO
0
1.1k
Featured
See All Featured
The Success of Rails: Ensuring Growth for the Next 100 Years
eileencodes
47
8k
Neural Spatial Audio Processing for Sound Field Analysis and Control
skoyamalab
0
250
Gemini Prompt Engineering: Practical Techniques for Tangible AI Outcomes
mfonobong
2
350
The SEO Collaboration Effect
kristinabergwall1
0
420
Imperfection Machines: The Place of Print at Facebook
scottboms
270
14k
Keith and Marios Guide to Fast Websites
keithpitt
413
23k
コードの90%をAIが書く世界で何が待っているのか / What awaits us in a world where 90% of the code is written by AI
rkaga
61
43k
Lightning talk: Run Django tests with GitHub Actions
sabderemane
0
160
WCS-LA-2024
lcolladotor
0
520
How Software Deployment tools have changed in the past 20 years
geshan
0
33k
CSS Pre-Processors: Stylus, Less & Sass
bermonpainter
360
30k
Why You Should Never Use an ORM
jnunemaker
PRO
61
9.8k
Transcript
Evaluation for classification dataitgirls3 Instructor Sunmi Yoon
Confusion Matrix
https://sumniya.tistory.com/26
Evaluation Metrics from Confusion Matrix
https://towardsdatascience.com/understanding-confusion-matrix-a9ad42dcfd62
Precision(ب), PPV(Positive Predictive Value) ݽ؛ TrueۄҊ ࠙ܨೠ Ѫ ী, पઁ
Trueੋ Ѫ ࠺ਯ Recall(അਯ), Sensitivity, hit rate पઁ True ী ݽ؛ True۽ ࠙ܨೠ ࠺ਯ “Precision݅ न҃ਸ ॳݶ ݽ؛ ੋ࢝೧Ҋ, Recall݅ न҃ॳݶ ݽ؛ ಌ” ܳ ࢤп೧ࠁࣁਃ.
Accuracy TP, TNਸ ݽف Ҋ۰ೞח . Label ࠛӐഋ बೡ ٸী
ࢎਊਸ ೧ঠ פ. F1 Score Precisionҗ Recall ઑചಣӐ Label ࠛӐഋ बೡ ٸী ݽ؛ ࢿמਸ ഛೞѱ ಣоೡ ࣻ णפ. Label ࠛӐഋ बೡ ٸী, Accuracyח ۽ࢲ न܉ࢿਸ णפ. ਬܳ ࢤп ೧ ࠁࣁਃ.
https://sumniya.tistory.com/26 ৵ ࣿಣӐ ইפҊ ઑചಣӐੋо?
ઑӘ݅ ؊ о ࠇद
https://towardsdatascience.com/understanding-confusion-matrix-a9ad42dcfd62 द Ӓܿਵ۽ جই৬ࢲ, ଘ ফܳ बਵ۽ ࢤп೮
https://towardsdatascience.com/understanding-confusion-matrix-a9ad42dcfd62 द Ӓܿਵ۽ جই৬ࢲ, ߣূ ফب э ࢤпೞݶࢲ ࠇद
(Әࠗఠ ഁтܾ ࣻ )
TRUE FALSE ࠙ܨѾҗ TRUE TP FP FALSE FN TN
TRUE FALSE ࠙ܨѾҗ TRUE TP FP FALSE FN TN
Precision Positive Predictive Value ࠙ܨ Ѿҗ(ݽ؛)ਸ बਵ۽
TRUE FALSE ࠙ܨѾҗ TRUE TP FP FALSE FN TN
Negative Predictive Value ࠙ܨ Ѿҗ(ݽ؛)ਸ बਵ۽
TRUE FALSE ࠙ܨѾҗ TRUE TP FP FALSE FN TN
Recall Sensitivity True Positive Rate ਸ बਵ۽
TRUE FALSE ࠙ܨѾҗ TRUE TP FP FALSE FN TN
ਸ बਵ۽ False Positive Rate
TRUE FALSE ࠙ܨѾҗ TRUE TP FP FALSE FN TN
ਸ बਵ۽ Specificity True Negative Rate
TRUE FALSE ࠙ܨѾҗ TRUE TP FP FALSE FN TN
ਸ बਵ۽ Fall-out rate False Positive Rate
https://towardsdatascience.com/understanding-confusion-matrix-a9ad42dcfd62 Ѧ ೞҊ ೮ભ. ߣূ ফب э ࢤпೞݶࢲ ࠇद (Әࠗఠ
ഁтܾ ࣻ )
TRUE FALSE ࠙ܨѾҗ TRUE TP FP FALSE FN TN
? TP ब ٜ ܻೞݶ, ?
TRUE FALSE ࠙ܨѾҗ TRUE TP FP FALSE FN TN
TN ब ٜ ? ܻೞݶ, ?
ഁтܻભ? ਗې Ӓ۠Ѣਃ
ӝୡח ೮ਵפө ઑӘ݅ ؊ ೧ ࠇद.
Confusion Matrix with Histogram
https://www.medcalc.org/manual/roc-curves.php Criterion, Threshold য়ܲଃ Distribution Actual True, ৽ଃ Actual False.
Threshold ਤ۽ח ݽف True۽ ஏೞח ݽ؛ Ҋ о೮ਸ ٸ,
https://www.medcalc.org/manual/roc-curves.php Thresholdܳ ӓױਵ۽ ஏ ز दெࠇद. যڃ ੌ ੌযաաਃ? Precision:
Recall: Specificity: Fall-out:
https://www.medcalc.org/manual/roc-curves.php Thresholdܳ ӓױਵ۽ ஏ ز दெࠇद. যڃ ੌ ੌযաաਃ? True
positive rate: True negative rate:
https://www.medcalc.org/manual/roc-curves.php ߣূ ߈۽ ز दெࠇद. যڃ ੌ ੌযաաਃ? True positive
rate: True negative rate:
Specificity৬ Sensitivity ҙ҅ https://www.medcalc.org/manual/roc-curves.php
ROC(Receiver Operating Characteristic) curve
рױೞѱח, Sensitivity৬ 1-Specificityܳ п ୷ਵ۽ ೞח 2ରਗ Ӓې https://www.medcalc.org/manual/roc-curves.php AUC
(Area Under Curve)
рױೞѱח, Sensitivity৬ 1-Specificityܳ п ୷ਵ۽ ೞח 2ରਗ Ӓې https://www.medcalc.org/manual/roc-curves.php Actual
True৬ Actual False distribution ৮߷ೞѱ эਸ ٸ (feature class ߸߹מ۱ হ) ROC curveח 45ب пب ࢶ
рױೞѱח, Sensitivity৬ 1-Specificityܳ п ୷ਵ۽ ೞח 2ରਗ Ӓې https://www.medcalc.org/manual/roc-curves.php Actual
True৬ Actual False distribution Ҁח হ ৮߷ೞѱ ܻ࠙ ؼ ٸ ROC ழ࠳ (feature class ߸߹ מ۱ ৮߷) ROC ழ࠳о ઝ࢚ױী оөࣻ۾ feature class ߸߹ מ۱ જҊ ೡ ࣻ .
ROC(Receiver Operating Characteristic) curve with Machine Learning
Classifierܳ ݅ٚח Ѥ, ف ѐ histogramਸ ӒܻҊ Thresholdܳ ೞח Ѫ
https://www.medcalc.org/manual/roc-curves.php
https://scikit-learn.org/stable/auto_examples/model_selection/plot_roc.html#sphx-glr-auto-examples-model-selection-plot-roc-py Histogramਸ Ӓ۷ח Ѥ ROC ழ࠳ܳ Ӓܾ ࣻ ח Ѫ!
https://scikit-learn.org/stable/auto_examples/model_selection/plot_roc.html#sphx-glr-auto-examples-model-selection-plot-roc-py ROC ழ࠳ܳ Ӓܾ ࣻ ח Ѥ ৈ۞ ROC ழ࠳
р ࠺Үܳ ా೧ જ ࢿמ ݽ؛ਸ ইյ ࣻ ח Ѫ!
AUCо = ݽ؛ ҅ೠ probabilityܳ ߄ఔਵ۽ Ӓܽ histogramٜ ੜ
ܻ࠙غয . = ݽ؛ Threshold(Decision BoundaryۄҊب ೠ)ী ؏ хೞ. = উੋ ஏਸ ೠ.
ݽ؛ ࢶఖী ROC ழ࠳ܳ ഝਊೠ = Decision Boundaryী ࢚ҙহ ؊
જ ݽ؛ਸ ח. = ganziо դ.
Ӓ۰ࠇद. ؘఠ: titanic ݽ؛ - sklearn.linear_model.LinearRegression - sklearn.linear_model.LogisticRegression -
sklearn.tree.DecisionTreeClassifier - sklearn.ensemble.RandomForestClassifier ١ whatever you want - Tree ҅ৌ ݽ؛ ҃ model predict_proba() ݫࣗ٘ܳ ࢎਊೞݶ ഛܫ ҅ ؾ פ. - ীח Thresholdܳ a ݅ఀ ز೧оݴ Sensitivity, Specificityܳ ҅೧ ઝܳ ҳೞ ࣁਃ. - যڌѱ ೞݶ Thresholdܳ ੜ زदఃݶࢲ ROC ઝܳ ନਸ ࣻ ਸөਃ? - ઝٜਸ ಣݶ࢚ী ନযࠁࣁਃ.
sklearn.metrics.roc_curve ܳ ഝਊ ೧ ࠇद. ؘఠ: titanic ݽ؛ - sklearn.linear_model.LinearRegression
- sklearn.linear_model.LogisticRegression - sklearn.tree.DecisionTreeClassifier - sklearn.ensemble.RandomForestClassifier ١ whatever you want ؊ աইоࢲ, - sklearnਸ ਊ೧ AUCب ҅ ೧ࠇद. - ৈ۞ ݽ؛ٜ ࢿמਸ ࠺Ү ೧ ࠇद. - DecisionTreeClassifierܳ ࢎਊ೮؊ۄب, ࢎਊೠ featureо ܰݶ ӒѤ ܲ ݽ؛ੑפ . - ఋఋץ ݈Ҋ, ܲ classification ޙઁীب ഝਊ೧ ࠁࣁਃ.