Why Gurugram Is Becoming a Hub for Machine Learning Training

Use-case framing
Decision: approve or hold a transaction
Prediction: probability the transaction is fraud
Action: hold if probability > threshold
Cost of a miss: fraud loss
Cost of a false alarm: annoyed genuine customer
Metric: recall at an acceptable false-alarm rate

Data inventory for a fraud project
Table          | Key fields                    | Freshness
transactions   | amount, merchant, time        | real time
devices        | device_id, new_device flag    | daily
customers      | account_age, past_disputes    | daily
labels         | is_fraud (confirmed later)    | delayed

# Serve a trained model with FastAPI
import joblib
from fastapi import FastAPI
from pydantic import BaseModel

model = joblib.load("fraud_model.joblib")
app = FastAPI()

class Txn(BaseModel):
    amount: float
    new_device: int
    account_age_days: int

@app.post("/score")
def score(t: Txn):
    prob = model.predict_proba([[t.amount, t.new_device, t.account_age_days]])[0][1]
    return {"fraud_probability": round(float(prob), 4)}

Role skill check (yes / not yet)
Data scientist: can I explain precision vs recall?      ___
ML engineer:    can I containerise a model API?         ___
Analytics eng:  can I write window functions in SQL?    ___
Product analyst: can I analyse an A/B test correctly?   ___

from sklearn.datasets import make_classification
from sklearn.model_selection import train_test_split
from sklearn.ensemble import RandomForestClassifier
from sklearn.metrics import classification_report

X, y = make_classification(n_samples=20000, weights=[0.98, 0.02], random_state=1)
X_tr, X_te, y_tr, y_te = train_test_split(X, y, stratify=y, random_state=1)
clf = RandomForestClassifier(n_estimators=200, class_weight="balanced", random_state=1)
clf.fit(X_tr, y_tr)
print(classification_report(y_te, clf.predict(X_te), digits=3))

import numpy as np
from sklearn.metrics import precision_score, recall_score

probs = clf.predict_proba(X_te)[:, 1]
for t in [0.2, 0.3, 0.5, 0.7]:
    pred = (probs >= t).astype(int)
    print(t, "precision", round(precision_score(y_te, pred), 3),
             "recall", round(recall_score(y_te, pred), 3))

Study week (working professional)
Mon  1 hr  statistics concept + exercise
Wed  1 hr  scikit-learn practice
Fri  1 hr  read one ML case study
Sat  3 hr  project work
Sun  1 hr  write-up and GitHub update

Project brief
Business problem: ______________
Decision supported: ______________
Data source: ______________
Baseline: ______________
Metric and why: ______________
Delivery: notebook / API / dashboard

Model card (short)
Model: random forest fraud classifier v1
Intended use: rank transactions for manual review
Training data: public synthetic dataset
Metrics: recall and precision at threshold 0.3
Limits: not tested on real payment data
Owner: ______________