Online Robotics Classes for Kids in BGC, Taguig (Metro Manila)

1. Introduction: Transitioning to Data Science

With the core programming modules completed, we begin a new chapter: Data Science. Data science is a foundational paradigm within Artificial Intelligence where we study data, transform it, and build predictive machine learning models.

To build any intelligent system, we must first understand what data actually is, how it is classified, and why it is the defining factor in artificial intelligence applications.

2. Data vs. Information: The Core Distinction

Understanding the transition from raw inputs to intelligent outputs requires distinguishing between data and information:

• What is Data? Data refers to raw facts, figures, and observations collected from various sources. In its raw form, data has not yet been processed or organized.
• What is Information / Knowledge? Information (or knowledge) is data that has been transformed and made meaningful. Transforming raw data allows us to extract insights and build predictive algorithms and automated systems.

When beginning any client engagement or engineering project, the workflow begins with domain understanding—learning the client's business context—followed immediately by extracting enough data to successfully train the target AI model.

3. The Three Types of Data

When collecting data from clients or external systems, the data arrives in three distinct forms:

1. Structured Data

• Characteristics: Highly organized data structured in well-defined tables with a fixed number of rows and columns.
• Common formats: Relational databases, CSV files (Comma Separated Values), and Microsoft Excel spreadsheets.
• Trade-off: It is the easiest data to visualize, clean, and process with algorithms. However, it is often the most difficult format for clients to prepare and provide.

2. Unstructured Data

• Characteristics: Data in the natural formats that humans process every day, lacking a predefined tabular layout.
• Common formats: Text documents, images, audio recordings, and videos.
• Trade-off: It is abundant and easy to collect from nature and everyday environments, but processing it with AI models is significantly more complex.

3. Semi-Structured Data

• Characteristics: Bridges the gap between structured and unstructured formats. It does not have fixed rows and columns like a spreadsheet, but its elements are clearly tagged and organized.
• Common formats: JSON (JavaScript Object Notation) and XML.
• Industry standard: JSON has become an industry standard for web services and APIs, outputting tagged records that can scale to extensive lengths while preserving structural clarity.

4. Why Data Matters in AI: "Garbage In, Garbage Out"

Data serves as the basic fuel for machine learning and AI models. The quality of your data directly dictates the quality and reliability of the resulting system:

• Garbage In, Garbage Out (GIGO): If an AI model is fed flawed, noisy, or low-quality data, it will inevitably produce poor, inaccurate outputs.
• Model limits: Even the most mathematically sophisticated and advanced AI model will fail if it is trained on poor data.
• Collecting high-quality data and formatting it properly is therefore the single most critical step in determining model accuracy and dependability.

5. The Four Pillars of High-Quality Data

To ensure an AI model trains effectively, the dataset must satisfy four fundamental criteria:

1. Relevance: The collected data must directly align with the specific problem being addressed. Clients often possess massive volumes of varied data across different departments, but software solutions address discrete problems one at a time. You must filter and gather only the data relevant to the active problem.
2. Accuracy: The data must be reliable and free from errors. While obtaining 100% clean data in real-world scenarios is virtually impossible, developers should target datasets that minimize errors (such as aiming for approximately 99% accuracy).
3. Diversity: The dataset must capture multiple variations, scenarios, and distinct use cases. If a dataset lacks variation, the model will overlook key requirements and fail when exposed to real-world edge cases.
4. Sufficiency: There must be enough data volume across every variation to properly train the AI architecture. Sufficient data ensures that the resulting model is robust and capable of generalizing accurately to unseen future examples.

6. The End-to-End Machine Learning Pipeline

Transforming client data into a functional AI application follows a structured, multi-stage engineering pipeline:

1. Data Collection: Gathering raw datasets from the client after acquiring an understanding of their domain.
2. Data Pre-processing:
   • Cleaning: Removing corrupt values, missing records, and anomalies.
   • Transformation: Converting the cleaned data into numerical and mathematical formats accepted by machine learning models.
3. Feature Engineering: Inspecting attributes and dropping irrelevant variables to keep the dataset precise and efficient.
4. Model Training: Iteratively training the machine learning model on the pre-processed data.
5. Model Evaluation: Testing the model against independent test data to evaluate performance, accuracy, and generalization before client deployment.
6. Deployment: Deploy the completed AI model to production.