Data Science Course Curriculum

Module 7: Machine Learning (Statistics, Visualization, ML, MLflow)

Topic 7.1: Exploratory Data Analysis (EDA) & Statistics

Theory:

• Types of variables (numerical, categorical)

• Descriptive statistics: mean, median, std dev, IQR

• Correlation and covariance

• Outlier detection

• Probability distributions and central limit theorem

  Lab:

• Use Pandas to explore and summarize customer transaction data

• Identify outliers in high-value transaction logs

  Scenarios:

• Detect suspicious activity based on statistical deviation

• Compare user spending patterns by region

  Tasks:

• Compute z-scores for transaction amounts

• Visualize missing values and fix them Challenges:

• Interpreting misleading correlations

• Handling skewed data distributions

Topic 7.2: Data Visualization

Theory:

• Plot types: bar, histogram, scatter, box, pie

• Visualizing distributions and relationships

• Dashboards using Plotly & Dash

• Effective storytelling with data

  Lab:

• Create matplotlib plots for ATM usage across states

• Build an interactive dashboard showing fraud trends

  Scenarios:

• Visualize time-series of withdrawals during weekends

• Highlight branch performance by customer type

  Tasks:

• Build scatter plot comparing income and loan default

• Create Dash app to filter customer data by risk score

  Challenges:

• Cluttered plots with large data

• Dynamic filtering and responsiveness in dashboards

Topic 7.3: Machine Learning (Supervised + Unsupervised)

Theory:

• Workflow: data → features → model → evaluation

• Classification (Logistic Regression, Decision Trees)

• Regression (Linear, Ridge, Lasso)

• Clustering (KMeans, DBSCAN)

• Model validation: train/test split, cross-validation

• Evaluation metrics: accuracy, precision, recall, F1

  Lab:

• Train a model to detect fraudulent transactions

• Predict loan default using classification

• Segment customers using KMeans

  Scenarios:

• Predict high-risk loan applicants

• Identify customer clusters for personalized offers

  Tasks:

• Create a pipeline: preprocessing → model → predict

• Tune hyperparameters using GridSearchCV

  Challenges:

• Imbalanced datasets causing false positives

• Feature leakage during training

• Overfitting and underfitting management

Topic 7.4: MLFlow for Model Lifecycle Management

Theory:

• MLflow Tracking: log parameters, metrics, models

• MLflow Projects: reproducible runs

• MLflow Models: deployment formats

• MLflow Registry: versioning and staging

Lab:

• Log model training runs in MLFlow

• Register fraud detection model

• Deploy latest version for real-time scoring

  Scenarios:

• Track accuracy across different fraud detection models

• Rollback to previous model version during failure

  Tasks:

• Log model predictions and accuracy to MLFlow

• Serve model using mlflow models serve

Challenges:

• Conflicts in model registry versions

• Tracking errors in distributed jobs

• Consistency in experiments and tags