A binary classification system that predicts whether a student will pass or fail before the exam, using behavioural and socioeconomic features. Designed to enable early intervention rather than post-exam remediation — where help can still make a difference.

Feature importance analysis revealed that study hours alone explains 44% of the predictive signal. This is not just a model that works — it is a model that points at something a teacher can act on. That is the difference between an ML experiment and a useful system. Logistic Regression outperformed Random Forest, suggesting the underlying decision boundary is predominantly linear — simpler models win when complexity is not warranted.

End-to-end business analytics on 149,116 real transactions across 3 NYC coffee shop locations covering January to June. Objective was to identify operational patterns, peak demand windows, and store-level performance differentials to surface actionable management insights.

Peak revenue at 9–10 AM across all 3 locations is directly actionable for staffing rotations. Hell's Kitchen outperforms comparable stores not in footfall but in average transaction value — a signal worth investigating for replication. The dashboard answers 7 distinct business questions without any additional analysis required.

Exploratory data analysis on 500+ retail orders across 5 product categories and 4 Indian metro regions. Core objective was identifying the source of profit leakage and building a visual system that communicates findings clearly enough to drive a policy recommendation.

A scatter plot of sales versus profit, coloured by discount level, revealed a clear loss cluster at high discount values. The correlation was strong enough to justify a direct policy recommendation. This is the kind of output that has business value — not just a model, but a finding that can be acted on in a meeting.