This repository focuces on building several Classification Machine Learning Models that predicts the decision of the loan approval based on customer attributes. I have worked on this project as part of Analytics Vidhya Hackathon
Although, I have a repository earlier with the same Hackathon- with PyCaret (using few lines of code)- This respository focuses on building several Classification models from scratch following the traditional Data Mining Approach of working on data preparation, data exploration, model buidling and validation
The code can be found in the LoanPrediction-ClassificationMLModels.ipynb" Notebook availabel in the same repository. "This repository includes the code for the below:
Technique 0.1: Pandas Profiling- A single step data exploratory technique
Technique 0.2: Checking the train data set to get feel of the dataset
Technique 0.3: To research on the variables (thier type and number of missing values)
Technique 0.4: Analyzing and researching the target variable for outliers, skewness
Technique 0.5- Analzying more about correlated features
Technique 0.6- Analyzing the top features that has close relationship/strong correlation with target. The top features are then checked for outlier removal (if needed)
Technique 0.7- Ground work for making data ready for data preparation
Technique 1: One Hot Encoding
Technique 2: Handling Missing data (Using Simple Random Imputation and Deterministic Imputation)
Technique 3: Feature Engineering
Technique 4: Removing correlated features
#4.1 Handling of constant columns with Variance Threshold
#4.2 Removing Quasi Constant features
#4.3 Removing duplicate features
#4.4 Removing correlated features using corr
Technique 5: Transformations and Scaling
Data Splitting Step 1: Dividing the train vs test separately after data preparation (as we combined them at the start before data preparation)
Data Splitting Step 2: Splitting into train vs validation (using train_test_split)
Model 1: Logistic Regression (that predicts the output in the form of probabilities between 0 and 1). Here probabilities refer to the chance of loan approval decision
Model 1.2: Logistic Regression that predicts output in the form of 0/1's as the Hackathon requires us to submit predictions in the form of Y/N
Model 2: XG Boost with Hyperparameters
Model 3: KNN Classifier
Model 4: Support Vector Machines (SVC)
Model 5: Naive Bayes Classifier
All the data files required to train/validate the model can be found in the data folder with in same repository and all the submission files that I have submitted for the Hackathon were included in the same repository under "submission_files" folder.
Hackathon link can be found under
https://datahack.analyticsvidhya.com/contest/practice-problem-loan-prediction-iii/
