Skip to content

📊 Chapter 9 — Data Science Process

A comprehensive overview of the end‑to‑end Data Science Process, from prior knowledge to deployment.

🔁 1. Overview

The Data Science Process transforms raw data into actionable insights through systematic stages:

graph TD
    A[Prior Knowledge] --> B[Data Preparation]
    B --> C[Modeling]
    C --> D[Evaluation]
    D --> E[Deployment]

Each stage iteratively refines understanding and improves decision quality.

🧠 2. Prior Knowledge

Definition

  • Domain expertise, business context, and problem framing.
  • Hypotheses and expectations derived from experience or literature.

Tasks

  • Identify the objective (classification, regression, clustering, etc.).
  • Define target variable(s).
  • Select metrics relevant to success.

Example

Suppose you’re building a loan default prediction model. Prior knowledge includes: - Customer demographics, credit history, and income behavior. - Bank risk policies. - Economic indicators.

graph LR
    A[Domain Knowledge] --> B[Feature Selection]
    A --> C[Metric Design]
    A --> D[Assumption Setting]

🧹 3. Data Preparation

Definition

Converting raw, messy data into a structured and analyzable form.

Sub‑steps

  1. Data Collection: from databases, APIs, logs, or files.
  2. Data Cleaning: handle missing values, outliers, and duplicates.
  3. Feature Engineering: derive new features, normalize, encode categorical variables.
  4. Splitting: train, validation, and test sets.
graph TD
    A[Raw Data] --> B[Cleaning]
    B --> C[Feature Engineering]
    C --> D[Splitting]
    D --> E[Modeling Ready Dataset]

Example (by hand)

Step Action Example
Missing Replace null income income = mean(income)
Outliers Cap extreme age values age = min(age, 80)
Encoding Convert gender male→1, female→0

🧮 4. Modeling

Definition

Selecting and training algorithms to represent data patterns.

Types of Models

Problem Model Family Example
Classification Logistic Regression, Decision Tree, SVM Predict customer churn
Regression Linear, Ridge, Lasso, Random Forest Predict house price
Clustering K‑Means, DBSCAN Group customers
Dimensionality Reduction PCA, t‑SNE Visualization 
graph LR
    A[Features] --> B[Train Model]
    B --> C[Evaluate on Validation Set]
    C --> D[Parameter Tuning]

Mathematical Foundation (example)

For linear regression:

\[ \hat{y} = \beta_0 + \sum_{j=1}^p \beta_j x_j \]
  • where \(\boldsymbol{\beta} = (X^TX)^{-1}X^Ty\)

📈 5. Evaluation

Purpose

Assess model performance and generalization.

Metrics

Type Metrics
Classification Accuracy, Precision, Recall, F1, ROC‑AUC
Regression RMSE, MAE, \(R^2\)
Clustering Silhouette Score, Davies‑Bouldin Index

Example Calculation

For predictions: \([0.9, 0.3, 0.8, 0.1]\), actual: \([1,0,1,0]\)
Threshold = 0.5 → Predictions = [1,0,1,0] → Accuracy = 100%.

graph TD
    A[Trained Model] --> B[Evaluation Metrics]
    B --> C[Compare Models]
    C --> D[Select Best Model]

🚀 6. Deployment

Definition

Integrating the trained model into a production environment.

Methods

  • Batch inference: periodic predictions (e.g., nightly).
  • Online inference: real‑time API endpoint.
  • Edge deployment: mobile or IoT model execution.
graph TD
    A[Best Model] --> B[API Service]
    B --> C[Application Integration]
    C --> D[User Feedback Loop]

Example Architecture

graph LR
    A[Model File #40; .pkl/.onnx #41;] --> B[Flask/FastAPI Service]
    B --> C[Docker Container]
    C --> D[Cloud Deployment <br> #40; AWS, Azure, GCP #41;]
    D --> E[Monitoring + Retraining]

🔄 7. Iteration and Continuous Learning

The process is cyclic — feedback from deployment improves earlier stages.

graph TD
    E[Deployment Feedback] --> D[Evaluation]
    D --> C[Modeling]
    C --> B[Data Preparation]
    B --> A[Prior Knowledge]

🧩 8. Summary Table

Stage Key Actions Tools
Prior Knowledge Define objective, assumptions Domain expertise, documentation
Data Preparation Cleaning, transformation Pandas, SQL, Excel
Modeling Algorithm training scikit‑learn, TensorFlow, PyTorch
Evaluation Performance metrics sklearn.metrics, visualization
Deployment Serve models Flask, FastAPI, Docker, AWS Sagemaker

🧠 9. Quick Exam Questions

  1. Explain how domain knowledge impacts feature engineering.
  2. What are the differences between batch and online deployment?
  3. Which metric is suitable for imbalanced classification?
  4. Describe a feedback loop in deployed systems.
  5. Write the formula for a linear regression model.