Predictive Modelling Notebooks

EDA and baseline models on four public datasets, from my first ML internship. Outputs shown exactly as the notebooks printed them.

The problem

Given a raw public dataset, can I clean it, find what drives the target and fit a baseline model, then report the result honestly, caveats included?

Internship exercises in exploratory analysis and predictive modelling: clean the data, find what drives the target, fit a baseline model.

Four public datasets: BigMart sales, credit card fraud, stock movement and Iris.
pandas, scikit-learn and seaborn in Jupyter notebooks.

01Checked missing values and duplicates, encoded categorical columns, and plotted distributions and correlations before modelling.
02Fraud: the dataset is 284,807 transactions with only 492 frauds (0.17%), so I balanced it by sampling 492 normal transactions to match.
03Stock: kept the 18 of 1,285 columns with no missing values, then scaled features before fitting.

Dataset	Model	Output	Measured
BigMart sales	Linear regression	R² 0.511	on the held-out test set
Credit card fraud	Logistic regression	94.5% accuracy	on the balanced training set; the notebook did not score the test set
Stock movement	Logistic regression	82.2% accuracy	on the test set; predictions are mostly one class, so compare against the majority-class rate
Iris	Logistic regression	97.3% accuracy	on the same data it was trained on

BigMart: item price (MRP) had the strongest correlation with outlet sales, 0.57.

The fraud accuracy is on the balanced training set; the notebook never scored the test set.
The stock model predicts mostly one class, so its 82.2% has to be read against the majority-class rate.
The Iris accuracy is measured on the data it was trained on, so it says nothing about new flowers.

I'm happy to walk through the decisions, what failed and what I'd change.