Bayes' theorem and applications — Hinglish notes

1.3.3 · AI-ML › Probability & Statistics

Bayes' Theorem Kya Hai?

YEH FORMULA KAAM KYUN KARTA HAI? Chalte hain ise scratch se derive karte hain.

First Principles Se Derivation

Conditional probability ki definition se shuru karo: $P (A ∣ B) = \frac{P ( A \cap B )}{P ( B )}$

KYUN? Yeh kehta hai: " $B$ diya hua $A$ ki probability" = "dono saath hote hain" divided by " $B$ kitni baar hota hai". Agar $B$ rare hai, toh $A$ aur $B$ ka saath aana relatively zyada significant hai.

Similarly, hum likh sakte hain: $P (B ∣ A) = \frac{P ( A \cap B )}{P ( A )}$

Is doosri equation ko rearrange karo: $P (A \cap B) = P (B ∣ A) \cdot P (A)$

KYUN? Hum joint probability ko express kar rahe hain "agar hum $A$ jaante hain toh $B$ kitna likely hai" times "pehli jagah $A$ kitna likely hai" ke terms mein.

Pehli equation mein substitute karo: $P (A ∣ B) = \frac{P ( B ∣ A ) \cdot P ( A )}{P ( B )}$

Yahi Bayes' theorem hai. Yeh sirf conditional probability definitions par algebra hai. Khoobsurati yeh hai: yeh conditioning ko reverse karta hai. Agar aap $P (B ∣ A)$ jaante ho lekin $P (A ∣ B)$ chahiye, toh Bayes aapko swap deta hai.

Denominator Ko Expand Karna

Aksar $P (B)$ directly pata nahi hota. Law of total probability use karo: $P (B) = P (B ∣ A) \cdot P (A) + P (B ∣\neg A) \cdot P (\neg A)$

KYUN? Event $B$ do mutually exclusive scenarios mein ho sakta hai: jab $A$ true hai ya jab $A$ false hai. Weighted probabilities ko sum karo.

Toh Bayes ban jaata hai: $P (A ∣ B) = \frac{P ( B ∣ A ) \cdot P ( A )}{P ( B ∣ A ) \cdot P ( A ) + P ( B ∣\neg A ) \cdot P ( \neg A )}$

Yeh form practice mein sabse zyada useful hai kyunki aap posterior ko sirf likelihoods aur priors se compute kar sakte ho.

Figure — Bayes' theorem and applications

Worked Examples

Worked example Example 1: Medical Diagnosis (Classic Application)

Ek disease 1% population ko affect karti hai. Ek test 95% accurate hai (true positive rate = 0.95, false positive rate = 0.05). Sawaal: Agar aap positive test karte ho, toh actually disease hone ki kya probability hai? Solution:

Maano $D$ = disease hai, $T$ = test positive
Prior: $P (D) = 0.01$ , $P (\neg D) = 0.99$
Likelihood: $P (T ∣ D) = 0.95$ (sensitivity), $P (T ∣\neg D) = 0.05$ (false positive rate) Bayes apply karo: $P (D ∣ T) = \frac{P ( T ∣ D ) \cdot P ( D )}{P ( T ∣ D ) \cdot P ( D ) + P ( T ∣\neg D ) \cdot P ( \neg D )}$ YEH STEP KYUN? Hum extended form use kar rahe hain kyunki $P (T)$ directly pata nahi, lekin hum ise do scenarios se compute kar sakte hain.

$P (D ∣ T) = \frac{0.95 \times 0.01}{0.95 \times 0.01 + 0.05 \times 0.99}$ $= \frac{0.0095}{0.0095 + 0.0495} = \frac{0.095}{0.059} \approx 0.161$

Jawab: Sirf 16.1% chance hai ki aapko disease hai!

ITNA KAM KYUN? Disease rare hai (low prior). Bhale hi test 95% accurate hai, lekin bahut zyada healthy log hain (99%) jo 5% false positive trigger kar sakte hain. False positives, true positives se zyada hote hain.

Bayes AI/ML Mein Powerful Kyun Hai

Probabilistic reasoning: Uncertainty ko explicitly handle karta hai (vs deterministic rules)
Prior knowledge incorporate karta hai: Agar pata hai ki disease rare hai, toh prior $P (D)$ low hai → confirm karna mushkil
Incrementally update karta hai: Aur evidence mila? Bayes phir se apply karo, purana posterior naya prior banega
Algorithms ka foundation: Naive Bayes, Bayesian networks, spam filters, medical AI, A/B testing

Common Mistakes

Connections

1.3.01-Conditional-probability-and-independence – Bayes conditional probability par built hai
1.3.02-Law-of-total-probability – Denominator mein $P (B)$ expand karne ke liye use hota hai
2.1.05-Naive-Bayes-classifier – Direct ML application
4.2.03-Bayesian-inference – Beliefs update karne ka philosophical framework
3.4.01-Maximum-likelihood-estimation – Frequentist alternative; MLE priors ignore karta hai
1.3.04-Random-variables-and-distributions – Bayes continuous distributions tak extend hota hai (Bayesian updating)

Recall Ek 12-Saal Ke Bacche Ko Explain Karo

Socho tum ek detective ho. Kisine cookies churaaye, aur tumhe chocolate crumbs milte hain. Tumhare suspects: tumhara bhai (jo chocolate bahut pasand karta hai, jab woh aaspaas hota hai toh 80% time cookies churata hai) aur tumhara dog (jo kuch bhi khaata hai, lekin sirf 20% chance hai ki woh crumbs chorega).

Crumbs dekhne se pehle, tumne socha tha ki tumhara bhai 40% time aaspaas hota hai (yahi tumhara prior hai). Crumbs evidence hain. Tumhara bhai agar cookies churata toh definitely chocolate crumbs choreta (high likelihood). Dog bhi chor sakta hai, lekin kam baar.

Bayes' theorem woh math hai jo tumhe batata hai: "Theek hai, tumne chocolate crumbs dekhe. Usse, ab kya NAYI chance hai ki yeh tumhara bhai tha?" Yeh tumhare starting guess ko combine karta hai is baat ke saath ki har suspect kitna achha un crumbs ko explain karta hai. Jo evidence create karne ki zyada likelihood hai, use bada update milta hai. Yahi Bayes hai!

#flashcards/ai-ml

Bayes' theorem ka formula kya hai? :: $P (A ∣ B) = \frac{P ( B ∣ A ) \cdot P ( A )}{P ( B )}$ jahan $P (A ∣ B)$ posterior hai, $P (B ∣ A)$ likelihood hai, $P (A)$ prior hai, $P (B)$ evidence hai.

Bayes' theorem ke char components kya hain? :: Posterior $P (A ∣ B)$ , Likelihood $P (B ∣ A)$ , Prior $P (A)$ , Evidence/Marginal Likelihood $P (B)$ .

Bayes' theorem mein denominator kyun zaroori hai?

Yeh posterior probability ko normalize karta hai taaki saare posteriors 1 tak sum hon aur probabilities 1 se exceed na karen.

Jab

P (B)

directly pata nahi ho toh use expand kaise karte hain?

Law of total probability use karo:

P (B) = P (B ∣ A) \cdot P (A) + P (B ∣\neg A) \cdot P (\neg A)

Medical test example mein, 95% test accuracy ke bawajood posterior probability sirf 16% kyun hai?

Kyunki disease rare hai (1% prior). Large healthy population ke false positives (99% × 5% = 4.95%) true positives (1% × 95% = 0.95%) se zyada hote hain.

Naive Bayes classifier mein "naive" ka kya matlab hai?

Yeh assume karta hai ki features class diya hua conditionally independent hain, jo usually galat hai lekin computation simplify karta hai aur aksar practice mein achha kaam karta hai.

P (A ∣ B)

aur

P (B ∣ A)

mein kya difference hai?

P (A ∣ B)

woh probability hai ki

B

occur hua given

A

ki (posterior), jabki

P (B ∣ A)

woh probability hai ki

B

occur hua given

A

occur hua (likelihood). Yeh generally equal nahi hote.

Bayes' theorem multiple hypotheses ko kaise handle karta hai?

P (A_{i} ∣ B) = \frac{P ( B ∣ A _{i} ) \cdot P ( A _{i} )}{\sum _{j = 1}^{n} P ( B ∣ A _{j} ) \cdot P ( A _{j} )}

jahan hypotheses

A_{i}

mutually exclusive aur exhaustive hain.

Bayes' theorem mein prior probability ka kya role hai?

Yeh evidence dekhne se pehle initial belief represent karta hai. Ek strong prior (bahut high ya low) ko significantly shift karne ke liye strong evidence chahiye.

Independence assumption violate hone ke bawajood Naive Bayes kaam kyun karta hai?

Kyunki posterior probabilities ka rank order aksar sahi hota hai bhale hi absolute values miscalibrated hon, jo classification ke liye sufficient hai.