Overfitting: rule ne data yaad kar liya, seekha kuch nahi.
Doosre naam: overfitting, curve fitting, data mining bias, multiple comparisons, p-hacking.
Ise kaise padhein
Jab tak rule chuna ja raha hai, tab tak har faisla saamne wale data ko dekh kar liya jaata hai. Threshold badalna, avadhi chunna, ek variation hataana — sab usi data par hote hain.
Uske baad wahi data yeh nahi bata sakta ki rule acha hai. Woh sirf yeh bata sakta hai ki rule us data par kitna fit baitha — aur woh toh banaya hi isliye gaya tha.
Yeh kya nahi batata
Overfitting rule mein nahi, use banane ke tareeke mein hoti hai. Isliye rule ko padh kar, ya uska tark sun kar, yeh tay nahi kiya ja sakta.
Ek overfitted rule ka tark bilkul samajhdaari bhara lag sakta hai. Ekmatra bharosemand jaanch yeh hai ki woh us data par kaisa karta hai jisne uske banne mein hissa nahi liya.
Jude hue concepts
Bahut baar dekhne se kuch na kuch mil hi jaata hai.
Bina kisi signal wale data par bhi, bahut se rules aazmane par unme se sabse achha achha hi dikhega — kyunki aapne bahut baar dekha.
Lesson kholo