• Data Dive: Loaded up a dataset with customer details—age, job type, income in Euros (converted from rupees), family size, and more. Cleaned it up, tossed duplicates, and peeked at patterns with charts.
• Stats Check: Ran some tests (like Chi-Square and Mann-Whitney U) to spot what drives insurance buys. Income and travel experience stood out big time.
• Model Building: Tested four models—Logistic Regression, Random Forest, Gradient Boosting, and SVM. Tweaked them with random search to boost recall (catching more buyers), then mixed them into an ensemble for fun.
• Results: Random Forest came out on top. Tuned version hit a solid ROC AUC of 0.70, and the recall-optimized one nailed 100% recall—catching every buyer, though with some extra noise.

• Data Prep: Started with 1,987 records, but 738 were repeats—yep, 37% duplicates. Dropped those, leaving 1,249 unique customers. Converted rupees to Euros (1 INR = 0.011 EUR) for a European vibe. No missing values, which was nice. Age ranged 25-35, income €3,300-€19,800.
• Exploration: Plotted histograms and bar charts. Age peaked around 28-29 and 32-33; income had clumps at €6,600 and €13,200. Most folks were private sector workers, grads, and not big flyers. Stats showed income, frequent flying, and overseas trips tied strongly to buying insurance.
• Model Run: First pass, Gradient Boosting led with 0.76 accuracy, but recall was meh (0.49). After tuning for recall, Random Forest shone—hit 100% recall with a 0.70 ROC AUC. The ensemble was decent too (0.68 ROC AUC). Trade-off? More false positives to catch everyone.
• Key Finds: Income's the star—higher earners buy more. Frequent flyers and globetrotters are in too. Age and family size matter a bit, but education and chronic diseases? Not so much.

• Data: Post-cleanup, 1,249 customers. Features like age (mean 29.8), family size (2-9), chronic diseases (33% yes), and insurance uptake (39% yes).
• Stats: Income difference was huge (p < 0.001, CI €11,381-€12,116 for buyers vs. €9,100-€9,608 for non-buyers). Frequent flyers (p=0.000009) and abroad travelers (p=10^-26) were way more likely to buy.
• Model Specs: Random Forest won after 50 tuning rounds—max_depth 12, 193 trees, balanced weights. Recall-optimized threshold dropped to 0.0188 for 100% recall. Precision took a hit (0.40), but that's the cost of not missing anyone.
• Challenges: Duplicates were a mess—37% is wild. Small sample after cleanup (1,249) limited some stats power. Recall focus meant more false positives.

• More data would help—1,249's a start, but thin for rare cases.
• Could tweak thresholds based on business goals—fewer false positives if precision matters more.
• Adding features like trip frequency or destination might sharpen it.

This model flags who's likely to buy travel insurance—great for targeting marketing without bugging everyone. Random Forest gives a solid balance, or the recall-optimized version catches all potentials if you're okay with extra follow-ups. It's not a crystal ball, but it's a handy tool for the company to boost sales smarter. Code's on my GitHub if you want a look!