데이터 마이닝 - Predictive Models

Predictive Modeling Applications

• Database marketing
• Financial risk management
• Fraud detection
• Process monitoring
• Pattern detection

Predictive Modeling Training data

Training data case

• categorical or numeric input and target measurements
• 범주형 or 수치형 입력변수와 목표변수 측정

Predictive Model

Predictive model

• a concise representation of the input and target association
• 입력변수와 목표변수 간 연관성의 간결한 표현

Predictions

• output of the predictive model given a set of input measurements
• 주어진 입력변수로부터 예측모형의 결과

Modeling Essentials

1. Predict new cases

• Three Prediction Types

  • decision predictions
    • A predictive model uses input measurements to make the best decision for each case
  • ranking predictions
    • A predictive model uses input measurements to optimally rank each case
  • estimate predictions
    • A predictive model uses input measurements to optimally estimate the target value

∴ Decision, Rank, Estimate

2. Select useful inputs

• Input Reduction Strategies
  • Redundancy
    • 불필요한 중복 제거하기
      → 같은 정보를 가진 변수가 있다면 하나만 남기고 제거
  • Irrelevancy
    • 무관성

∴ Eradicate redundancies and irrelevancies (불필요한 중복과 무관성을 제거하라)

Why dimension reduction?

• Avoid curse of dimensionality
• Reduce amount of time and memory required by data mining algorithms
• Allow data to be more easily visualized
• Get insight easier
• May help to eliminate irrelevant features or reduce noise
• Reduce complexity of models

The curse of dimensionality (차원의 저주)

• Refers to the problems associated with multivariate data analysis as the dimensionality increases
• “A function defined in high-dimensional space is likely to be much more complex than a function defined in a lower-dimensional space, and those complications are harder to discern” –Friedman
• This means that a more complex target function requires denser sample points to learn it well!

ex) 각 변수가 0~1 사이에 분포하고, 변수(차원)이 10개일 때 전체 자료 중 5%를 관찰할려면 변수별 필요한 범위 $x$는?
$x^{10}$ = 0.05 → 필요한 한 변의 길이는 0.74

ex) 변수가 30개라면? $x^{30}$ = 0.05 → 필요한 한 변의 길이는 0.86

∴ 차원이 늘어남에 따라 필요한 데이터 양이 급격히 증가
↔ 같은 비율(%)의 공간을 설명하기 위해 필요한 데이터 양이 급격히 증가

∴ 변수가 많아도 전체 중 관찰할 수 있는 부분은 극소수
→ 변수가 증가할수록 심각해짐

∴ 데이터의 양보단 질이 중요

3. Optimize complexity

1. Data Partitioning
   • Partition available data into training and validation sets
2. Predictive Model Sequence
   • Create a sequence of models with increasing complexity
3. Model Performance Assessment
   • Rate model performance using validation data
4. Model Selection
   • Select the simplest model with the highest validation assessment

∴ Tune models with validation data