Skip to content

Latest commit

 

History

History
181 lines (108 loc) · 6.04 KB

README.md

File metadata and controls

181 lines (108 loc) · 6.04 KB
Raw

Adult-Census-Income

Purpose:

This project is to predict a person's salary lies in either 50K+ or 50K-.

1.1 Data Extraction

  • The Adult-Census-Income is from kaggle:

    https://www.kaggle.com/uciml/adult-census-income

  • Dataset Features:

    Salary, age, workclass, fnlwgt, education, education_num, marital-status, occupation, relationship, race, sex, capital-gain,capital-Loss, hours-per-week, native-country

  • The Salary Feature is the label we want to predict.

1.2 Data Preprocess

  • Transforming Feature

    Some Features contains string values, which can't be calculated in numpy library.
Convert categorical values into numerical values.

    Method: Label Encoding, One-Hot Encoding

  • Imputing Missing Values

    Some Features contain the missing values like NAN, ? or space.
Imputate missing value via feature mode.

    Method: Feature Mode, Median and Mean

  • Dealing with Imbalanced Data

    Salary Feature is unbalanced labeled. 
The salary > 50k is around 80% and <= 50K is 20%.

    Method: Bagging and Undersampling

1.3 Model Selection

  • Random Sampling of Training Data

    From training dataset, use Undersampling method by selecting a subset of the majority examples to match the number of minority examples to create a balanced dataset.

  • Building Classifier

    Classification Algorithms Selected:

    K-nearest-neighborhood, Support Vector Machine, Logistic Regression, Random Forest, Navie Bayes, Decision Tree, Adaboost Decision Tree

  • Ensemble Learning

    Each instance of training and test data is classified 0 (corresponding to less or equal than 50K dollars annually) or 1 (corresponding to greater than 50K annually) using the learned classifiers.

    Final prediction is made by taking a majority vote model among the predictions of these classifiers

1.4 Result

  • Bagging Decision tree:

    Confusion matrix:

    Prediction Truth Truth
    <=50K >50K
    <=50K 0.63 0.13
    >50K 0.05 0.19

    Prediction accuracy for instances label <= 50K is 0.83.

    Prediction accuracy for instances label > 50K is 0.80.

    Overall Test Accuracy is 0.82.

  • Random Forest

    Confusion matrix:

    Prediction Truth Truth
    <=50K >50K
    <=50K 0.62 0.15
    >50K 0.04 0.20

    Prediction accuracy for instances label <= 50K is 0.81.

    Prediction accuracy for instances label > 50K is 0.85.

    Overall Test Accuracy is 0.82.

  • Logistic regression

    Confusion matrix:

    Prediction Truth Truth
    <=50K >50K
    <=50K 0.64 0.12
    >50K 0.04 0.19

    Prediction accuracy for instances label <= 50K is 0.84.

    Prediction accuracy for instances label > 50K is 0.82.

    Overall Test Accuracy is 0.83.

  • K-Neighbor

    Confusion matrix:

    Prediction Truth Truth
    <=50K >50K
    <=50K 0.61 0.15
    >50K 0.05 0.19

    Prediction accuracy for instances label <= 50K is 0.80.

    Prediction accuracy for instances label > 50K is 0.80.

    Overall Test Accuracy is 0.80.

  • Support Vector Machine(SVM)

    Confusion matrix:

    Prediction Truth Truth
    <=50K >50K
    <=50K 0.63 0.13
    >50K 0.05 0.19

    Prediction accuracy for instances label <= 50K is 0.83.

    Prediction accuracy for instances label > 50K is 0.80.

    Overall Test Accuracy is 0.82.

  • Naïve Bayes

    Confusion matrix:

    Prediction Truth Truth
    <=50K >50K
    <=50K 0.61 0.15
    >50K 0.04 0.20

    Prediction accuracy for instances label <= 50K is 0.80.

    Prediction accuracy for instances label > 50K is 0.84.

    Overall Test Accuracy is 0.81.

  • Ensemble: Majority Vote for 7 Learned Classifiers

    Confusion matrix:

    Prediction Truth Truth
    <=50K >50K
    <=50K 0.63 0.15
    >50K 0.04 0.20

    Prediction accuracy for instances label <= 50K is 0.82.

    Prediction accuracy for instances label > 50K is 0.84.

    Overall Test Accuracy is 0.83.