Modelni baholash

import numpy as np
from sklearn.metrics import accuracy_score

# 1000 ta bemordan 20 tasi kasal
haqiqat = np.array([0] * 980 + [1] * 20)

# "Hech kim kasal emas" deydigan model
bashorat = np.zeros(1000)

print(f"Aniqlik: {accuracy_score(haqiqat, bashorat):.3f}")

from sklearn.metrics import confusion_matrix, ConfusionMatrixDisplay

matritsa = confusion_matrix(y_sinov, bashorat)
print(matritsa)

import matplotlib.pyplot as plt

ConfusionMatrixDisplay(matritsa, display_labels=["Sog'lom", "Kasal"]).plot(cmap="Blues")
plt.show()

from sklearn.metrics import precision_score, recall_score, f1_score

print(f"Precision: {precision_score(haqiqat, bashorat):.3f}")
print(f"Recall:    {recall_score(haqiqat, bashorat):.3f}")
print(f"F1:        {f1_score(haqiqat, bashorat):.3f}")

from sklearn.metrics import classification_report

print(classification_report(
    y_sinov, bashorat,
    target_names=["Sog'lom", "Kasal"],
))

ehtimollik = model.predict_proba(X_sinov)[:, 1]

print(f"{'Chegara':>8} {'Precision':>10} {'Recall':>8} {'F1':>7}")

for chegara in [0.1, 0.3, 0.5, 0.7, 0.9]:
    b = (ehtimollik >= chegara).astype(int)
    print(f"{chegara:>8.1f} "
          f"{precision_score(y_sinov, b, zero_division=0):>10.3f} "
          f"{recall_score(y_sinov, b):>8.3f} "
          f"{f1_score(y_sinov, b):>7.3f}")

from sklearn.metrics import roc_curve, roc_auc_score, RocCurveDisplay

fpr, tpr, chegaralar = roc_curve(y_sinov, ehtimollik)
auc = roc_auc_score(y_sinov, ehtimollik)

print(f"AUC: {auc:.3f}")

from sklearn.metrics import (
    mean_squared_error, mean_absolute_error,
    r2_score, mean_absolute_percentage_error,
)

print(f"MSE:   {mean_squared_error(y_s, b):.1f}")
print(f"RMSE:  {np.sqrt(mean_squared_error(y_s, b)):.1f}")
print(f"MAE:   {mean_absolute_error(y_s, b):.1f}")
print(f"MAPE:  {mean_absolute_percentage_error(y_s, b) * 100:.1f}%")
print(f"R2:    {r2_score(y_s, b):.3f}")

from sklearn.linear_model import LogisticRegression

# 1. Sinf vaznlari
model = LogisticRegression(class_weight="balanced")

# 2. Qo'lda vazn berish
model = LogisticRegression(class_weight={0: 1, 1: 50})

# 3. Chegarani o'zgartirish
bashorat = (model.predict_proba(X_s)[:, 1] >= 0.3).astype(int)

# 4. Namunalar sonini o'zgartirish (imbalanced-learn kutubxonasi)
# pip install imbalanced-learn
from imblearn.over_sampling import SMOTE

smote = SMOTE(random_state=42)
X_muvozanat, y_muvozanat = smote.fit_resample(X_oqitish, y_oqitish)

print(Counter(y_oqitish))
print(Counter(y_muvozanat))

from sklearn.dummy import DummyClassifier, DummyRegressor

# Har doim eng ko'p uchraydigan sinfni aytadi
asos = DummyClassifier(strategy="most_frequent")
asos.fit(X_o, y_o)

print(f"Asos model:  {asos.score(X_s, y_s):.3f}")
print(f"Sizning model: {model.score(X_s, y_s):.3f}")