report.txt

- SHIVANI THAKKAR (sdt170030)
- NIDHI VAISHNAV (ntv170030)

Preprocessing 

1) In preprocessing, we have removed the missing values first. Then we have converted all categorical/nominal attributes to
   numerical attributes. After that all attributes are normalized.


Neural Network implementation

1) Neural Network implementation takes into account various parameters like the training testing split, number of hidden
   layers, number of neurons in each hidden layer, number of iterations and the learning rate.

2) By testing neural network on given three datasets, we observed that if number of iterations increases, it leads to low
   mean square error. Number of hidden layers also plays an important role. If there are many hidden layers, itt leads to 
   overfitting. So number of hidden layers should be intermediate. If learning rate is too low, it takes long time to 
   converge. If learning rate is too high, accuracy gets affected. We have kept intermediate learning rate.


The results are demonstrated in the screenshots attached.