A collection of five end-to-end data science projects covering exploratory data analysis, classification, and regression — implemented in Python using real-world datasets.

• Language: Python 3.10+
• Data Manipulation: pandas, numpy
• Visualization: matplotlib, seaborn
• Machine Learning: scikit-learn
• Environment: Jupyter Notebook

📦 Data-Analytics-Internship/
├── 📓 task1_iris_visualization.ipynb
├── 📓 task2_credit_risk_prediction.ipynb
├── 📓 task3_customer_churn_prediction.ipynb
├── 📓 task4_insurance_claim_prediction.ipynb
├── 📓 task5_loan_acceptance_prediction.ipynb
├── 📄 task2_dataset.csv
├── 📄 task3_dataset.csv
├── 📄 task4_dataset.csv
├── 📄 task5_dataset.csv
└── 📄 README.md

Notebook: task1_iris_visualization.ipynb

Objective: Explore and visualize the classic Iris dataset to understand feature distributions and inter-species relationships.

Highlights:

• Loaded and inspected dataset using .shape, .columns, and .head()
• Built scatter plots, histograms, and box plots to analyze feature distributions
• Generated a full pair plot and correlation heatmap
• Trained a KNN classifier as a baseline — confirmed strong species separability

Key Insight: Petal length and petal width are far more discriminative than sepal features. Setosa is linearly separable from the other two species.

Skills Demonstrated: Data loading & inspection · Visualization with matplotlib & seaborn · Basic classification

Notebook: task2_credit_risk_prediction.ipynb
Dataset: task2_dataset.csv — 367 loan applicants (Kaggle Loan Prediction Dataset)

Objective: Predict whether a loan applicant has good or bad credit history as a proxy for default risk.

Highlights:

• Handled missing values in 6 columns using mode/median imputation
• Encoded categorical features (Gender, Education, Property_Area, etc.) with Label Encoding
• Visualized loan amount, income, and credit distributions with histograms and box plots
• Compared Logistic Regression and Decision Tree classifiers
• Evaluated using accuracy, confusion matrix, ROC-AUC, and classification report

Key Insight: Credit_History, Property_Area, and LoanAmount are the strongest predictors of credit risk. Semiurban applicants and graduates show higher good-credit rates.

Skills Demonstrated: Missing value imputation · Categorical encoding · Binary classification · Model evaluation

Notebook: task3_customer_churn_prediction.ipynb
Dataset: task3_dataset.csv — 10,000 bank customers (Churn Modelling Dataset)

Objective: Identify bank customers likely to leave (churn) based on their profile and account behaviour.

Highlights:

• Dropped non-predictive identifier columns (RowNumber, CustomerId, Surname)
• Applied Label Encoding for Gender and One-Hot Encoding for Geography
• Analyzed churn by age, balance, number of products, active membership, and geography
• Trained a Random Forest classifier with feature importance analysis
• Validated with 5-fold cross-validation and ROC curve

Key Insight: Age, balance, and number of products are top churn drivers. Customers with 3+ products and inactive members churn at significantly higher rates. German customers show the highest churn rate by geography.

Skills Demonstrated: Categorical encoding (Label + One-Hot) · Feature importance analysis · Random Forest · Cross-validation

Notebook: task4_insurance_claim_prediction.ipynb
Dataset: task4_dataset.csv — 1,338 policyholders (Medical Cost Personal Dataset)

Objective: Estimate individual medical insurance charges based on personal and lifestyle attributes.

Highlights:

• Explored charge distributions (raw and log-transformed)
• Visualized the impact of BMI, age, and smoking status on charges with scatter plots and box plots
• Trained and compared Linear Regression and Random Forest Regressor
• Evaluated models using MAE, RMSE, and R²
• Plotted residuals to assess model fit

Key Insight: Smoking is the single most powerful predictor — smokers are charged 3× more on average. BMI and smoking interact strongly: high-BMI smokers face the highest charges. Linear Regression shows clear heteroscedasticity, while Random Forest handles non-linearities better.

Skills Demonstrated: Regression modeling · MAE & RMSE evaluation · Residual analysis · Feature importance

Notebook: task5_loan_acceptance_prediction.ipynb
Dataset: task5_dataset.csv — 41,188 customers (UCI Bank Marketing Dataset)

Objective: Predict which bank customers are likely to accept a personal loan / term deposit offer from a marketing campaign.

Highlights:

• Loaded semicolon-delimited UCI dataset with 21 features including economic indicators
• Replaced 'unknown' string values with NaN and imputed with mode
• Encoded all categorical variables with Label Encoding
• Performed EDA on age, call duration, job type, marital status, and campaign contacts
• Trained Logistic Regression and Decision Tree models
• Evaluated with accuracy, ROC-AUC, confusion matrix, and 5-fold cross-validation

Key Insight: Call duration and economic indicators (euribor3m, nr.employed) are the strongest predictors. Previous campaign successes are highly predictive of future acceptance. The dataset has an ~11.3% acceptance rate, making ROC-AUC the critical evaluation metric.

Skills Demonstrated: Large dataset handling · Feature encoding · Imbalanced classification · Cross-validation · ROC analysis

pip install pandas numpy matplotlib seaborn scikit-learn jupyter

1. Clone this repository:

   git clone https://github.com/MUsman054/Data-Analytics-Internship.git
   cd Data-Analytics-Internship

2. Launch Jupyter Notebook:

   jupyter notebook

3. Open any .ipynb file and run all cells (Cell → Run All).

Note: Each notebook loads its dataset from a CSV file in the same directory. Make sure the .csv files are present alongside the notebooks before running.

This portfolio was built as part of a hands-on data science learning journey, demonstrating practical skills in EDA, machine learning, and model evaluation using real-world datasets.