Neural Network

An Artificial Neural Network (ANN) is an information processing paradigm that is inspired by the way biological nervous systems, such as the brain, process information.  It is composed of a large number of highly interconnected processing elements (neurones) working in unison to solve specific problems. ANNs, like people, learn by example. An ANN is configured for a specific application, such as pattern recognition or data classification, through a learning process.

Before dropping finishing column

Final look of the dataframe

We divided our data set as “X” which contained all the columns except “forvet” column and “y” which only contained forvet column. After that we applied 70% train – test split.

Version #1

Again 70% train-test split size was used thus we got 90 ( 300*(30/100) ) values (size of y_test) in our predicted array.

Predicted vs Actual Forvet Values

Results of the dataframe with 300 row

Accuracy Score : 93%

Version #2

For this version, we reduced number of features in the dataframe and applied the NN algorithm and results were suprising.

The dataset used for this version

Scatter Plot of actual and predicted data

Accuracy Score: 69%

Version #3

After training and fitting our model we got an accuracy score.

Accuracy Score 89%.

Accuracy of Neural Networks

 

Multilayer Perceptron

We used our initial dataset (the one with 17994 rows).

Here initially we started with 40 iterations but we got a warning saying “maximum iterations reached and the optimization has not converged yet” so, we increased the iteration number to 100.

 

Loss per iteration

Iteration stop, because no decrease in loss

 

Training set score and loss

Here, in the graph we can easily see the loss per iteration. At first there is a jump from 0.40’s to 0.50 and then there is a sudden decrease to 0.20’s. After that very small decrease in the loss is observed, so after some point where there is no more decrease, iterations are stopped.