Loan-Prediction-System

File named Data Insights

Task 1

1. Do the Exploratory Data Analysis & share the insights.

Firstly read both csv files and did the exploratory analysis of both checked the relationship between the Number of existing loans at this bank with High Risk applicant and bivariate analysis of Principal_loan_amount and Balance_in_existing_bank_account_(lower_limit_of_bucket)

2. How would you segment customers based on their risk (of default)?

   Merged the Applicant and Loan dataframes then removed the non essential columns merged
   Applicant Id and High risk applicant column in same table ordered by their risk.

3. Which of these segments / sub-segments would you propose be approved?

   A person having more Number_of_existing_loans_at_this_bank and Loan_history should have low chance of paying back loan

4. Tell us what your observations were on the data itself (completeness, skews).

   The data had a lot of Non essential column, missing values and maximum categorical variables

TASK-2

File Named - ML model

1 . Explain your intuition behind the features used for modeling.Are you creating new derived features? If yes explain the intuition behind them

I used a decision tree because its a flowchart-like structure in which each internal node represents a "test" on an attribute (e.g. whether a coin toss comes up heads or tails), each branch represents the outcome of the test, and each leaf node represents a class label (decision taken after computing all attributes). It is useful in cases where data set is very scaterred in both X and Y axis in graph and in which multiple attributes are responsible for prediction. After that I used GaussianNB because It is easy and fast to predict the class of the test data set. It also performs well in multi-class prediction. And finally a Logistic regression

2. Are there missing values? If yes how you plan to handle it.

   Yes there were missing values I handled it by using Pandas methods help us achieve this in a single line of code for every column. The fillna()method fills the empty fields with whatever parameter is given.Calculating the mean or mode of this array of values, and passing it to fillna() completes this step.Using mode works best in our case, as most columns are binary. Moreover, Mode will simply put the most occurring instance in place of empty fields, which, under the circumstances, would be the best guess.

3. How categorical features are handled for modeling.

   Using use the get_dummies function of Pandas for non-binary values like 'Property' to automatically one-hot encode them

4. Describe the features correlation using correlation matrix. Tell us about few correlated feature & share your understanding on why they are correlated.

   A correlation matrix is a table showing the correlation coefficients between different sets of variables

   I've used seaborn heatmap for correlation in which the red/warm color denotes the stronger corelation and in our dataframe Principal_loan_amount and months_loan_taken for a strong corelation because if the we increase loan amount the duration of loan also increases and Emi rate and loan amount have inverse corelation

5. Do you plan to drop the correlated feature? If yes then how.

   Yes I plan to showcase this feature using Seaborn heatmap because coolwarm colors really makes it easy to identify the correlation in dataframe.

6. Which ML algorithm you plan to use for modeling. and Their different parameters

   I used a decision tree because its useful us in cases where data set is very scaterred in both X and Y axis in graphs. It gave a accuracy of 64%

   After That I used GaussianNB It perform well in case of categorical input variables compared to numerical variable. I performed it to compare it with the Logistic Regression and see the difference between the accuracy

   Finally I used the Logistic Regression because it works well with bonary numeric values since after normalising most of the values in numeric binary Logistic regression gave us the best accurancy of 72%.

7. Explain how you will export the trained models & deploy it for prediction in production.

   We can use TensorFlow model creation then create webservice to serve the client request create deployment container with webservice, test the container, put it in production