This project was the second assignment of the Computational Intelligence course at Shahid Beheshti University. The aim of the project was to take advatage of the SVM(using scikit-learn) and KNN(from scratch and without other external libraries) in order to predict the income of the citizens whether they exceed from some constant number or not.
The dataset has been obtained from KEEL. The Adult data set was extracted in 1994 from census data of the United States. It contains continuous and nominal attributes, describing some social information (age, race, sex, marital status, ...) about the citizens registered.training neural networks in PyTorch)
The dataset itself needs lots of data-cleaning! There are many missing values for features which should be filled by existing techniques like Imputation or Removing.
Also the one-hot encoding technique has been used for categorical features. Binning method also was used for some features like Age.
At last, standardization and normalization was used for better convergence of the model.
SVM with differenet kernels like RBF and Polynomial has been tried and the results are plotted in the notebook. Also there is an efficient implementation of KNN in the notebook. Other explanation of the this procedure is documented in the citizenIncomeExceedingPrediction.ipynb.