Video summary

Data Mining : Perbedaan Pengelompokan, Klasifikasi dan Prediksi (Data Supervised dan Unsupervised)

Main summary

Key takeaways

Educational

Main ideas / concepts covered

  • Introduction to the speaker and their background

    • The presenter introduces themselves as Yoga Religia, sharing their work/education background and current focus areas: data science, data mining analysis, digital marketing, and e-commerce.

  • What Data Mining means (and common misconceptions)

    • Data mining is described as a method to discover hidden knowledge in data stored in a database.
    • It can be understood as a series of processes to extract added value/knowledge that would be hard to obtain manually.
    • The goal is to find relationships, patterns, and useful indications from existing data.

  • What data mining does

    • Performs automatic or semi-automatic analysis on large volumes of incoming data to extract previously unknown patterns, such as:
      • clusters
      • anomalies
      • associations

Data Mining process: 6 phases (Larose, 2016)

The video states that data mining should follow six sequential phases, because each phase determines what methods can be used next and helps ensure the business problem is actually solved.

  1. Business Understanding

    • Understand the business process and manual workflow.
    • Identify the business problem the organization wants to solve using data mining.
    • Example (banking): determine which customers should be offered loans and which should not.

  2. Data Understanding

    • Determine what data is suitable to solve the identified business problem.
    • Assess the data type and suitability (e.g., what the data is, size, whether it can be used).

  3. Data Processing (Preprocessing)

    • Clean and prepare the data by handling:
      • errors/noise
      • missing gaps
    • Goal: make data “normal” and error-free so it yields good performance.
    • If data is already good: just check it; if not: parse/clean it so it’s usable.

  4. Modeling

    • Use prepared data to build models to extract hidden information.
    • The video frames modeling as producing “results/insight behind the data.”

  5. Evaluation

    • Assess the performance of the model/results.
    • Emphasizes that without evaluation, a solution may perform poorly and fail to solve (or even worsen) the business issue.
    • Purpose: measure whether the model is good enough to proceed.

  6. Deployment / Implementation

    • If evaluation shows good performance, implement the solution to address the business problem.
    • The video stresses that the phases should be followed in order.

Modeling methods in data mining (6 types)

The video lists six modeling types, describing what each one is used for:

  1. Association

    • Finds associative rules between combinations of items.
    • Example: if a customer buys milk, what other items are likely also purchased.
    • Goal: improve grouping/arrangement of products that commonly occur together.

  2. Regression

    • Models the relationship between dependent and independent variables.
    • Example: MSME adoption influenced by technological, organizational, and environmental factors.

  3. Clustering

    • Groups data based on similar characteristics.
    • Used when you need to cluster data that otherwise has no predefined group.
    • The video describes clustering as organizing data to make it easier to understand.

  4. Classification

    • Assigns data to known categories based on relationships to sample/labelled data.
    • The video earlier ties classification to the presence of labels (supervised learning).

  5. Prediction

    • Predicts values/groups for new/unseen data when the group label is not yet known.
    • The video says it’s similar to classification, but focuses on the different data setup.

  6. Forecasting (time-based prediction)

    • Uses relationships from previous cycles/time-series data to predict future values.
    • Example: if the pattern continues this year, what might it be like next year.

The presenter mentions they will later focus (on their channel) on clustering, classification, and prediction because many people are confused about these.

Supervised vs Unsupervised vs Semi-supervised learning (3 data types)

The video explains three data types used for modeling:

  • Supervised learning

    • Each data item has a label/class.
    • Example conceptually: data includes attributes and a known category (e.g., thin/fat).

  • Unsupervised learning

    • No labels/classes are provided—only attributes exist.
    • The model must discover structure/groups automatically.

  • Semi-supervised learning

    • A combination of labelled and unlabelled data.
    • Example described:
      • some data points are labelled (e.g., thin/fat),
      • others have only attributes and no class.

How these map to the three featured tasks

  • Clustering → uses unsupervised learning (data without labels)
  • Classification → uses supervised learning (labelled training data)
  • Prediction → described as using a semi-supervised setup:
    • training includes labelled information,
    • testing/unlabelled portions require assigning classes using the learned model.

Detailed methodology/instructions for classification (as described)

The video provides a practical process for classification models:

  • Step 1: Split labeled data

    • Divide data into:
      • training data
      • testing data

  • Step 2: Validation approach

    • The video describes dividing data using fold validation:
      • Cross validation
        • Data division is done randomly.
      • Split validation
        • Specify a percentage split, typically:
          • at least 75% training
          • remaining 25% testing (or similar ratios)

  • Step 3: Train model

    • Use the training set with a classification algorithm (examples listed: Naive Bayes, KNN, Decision Tree, SVM, Random Forest, Neural Network, etc.).

  • Step 4: Test model

    • Apply the trained model to the testing set.

  • Step 5: Measure performance

    • Metrics mentioned:
      • Accuracy (how often predictions match correctly)
      • Precision (correctness relative to predicted positives)
      • Recall (how well actual positives are captured)
    • Mentions using a confusion matrix concept to derive/understand these metrics.

Clustering modeling details (as described)

  • Clustering aims to group data into categories (e.g., thin/fat in the presenter’s example) based on similarity.

  • The video states there are two common clustering approaches:

    • Partitional clustering
    • Hierarchical clustering (referred to as “direct” clustering / hierarchy in the subtitles)

  • Examples of algorithms mentioned for clustering (partitional):

    • k-means and related algorithms (subtitles contained some garbled text, but the intent is to reference common k-means-style clustering approaches).

  • The presenter notes that “which is better” depends on understanding the business process.

Speakers / sources featured

  • Speaker (presenter): Yoga Religia
  • Source cited: Larose (2016) for the six-phase data mining process

Original video