Building a Neural Network For Classification

5.10. Building a Neural Network For Classification#

For this example we’ll use our pass/fail dataset. A 1 means the student passed the exam, and a 0 means they failed. pass_fail.csv.

import pandas as pd

data = pd.read_csv("pass_fail.csv")
print(data)

Building a neural network for classification is very similar to building a neural network regression, the main differences is that instead of importing MLPRegressor we use MLPClassifier.

from sklearn.neural_network import MLPClassifier

Then when we make predictions, we can get the raw class predictions using .predict or we can obtain the probabilities using .predict_proba.

Here is a complete example.

from sklearn.neural_network import MLPClassifier
import pandas as pd
import numpy as np

data = pd.read_csv("pass_fail.csv")
x = data[["Time Spent Studying (hours)"]].to_numpy()
y = data["Exam Result"].to_numpy()

nn = MLPClassifier(
    hidden_layer_sizes=(2), max_iter=500, learning_rate_init=0.1, random_state=0
)
nn.fit(x, y)

x_test = np.array([[0], [3], [5]])
print(nn.predict_proba(x_test).round(4))  # Round to 4 decimal places
print(nn.predict(x_test))

Let’s interpret the outputs.

The 3 test samples we provide are of students who have studied 0 hours, 3 hours and 5 hours respectively. The output probabilities are:

[[1.     0.    ]
[0.9839 0.0161]
[0.     1.    ]]
[0 0 1]

Here is a summary of what this means:


Student 1	0	1	0
Studetn 2	3	0.9839	0.0161
Student 3	5	0	1

Based on these probabilities students 1 and 2 are predicted to fail and student 3 is predicted to pass. This corresponds to the predictions we see:

[0 0 1]


Student 1	0	1	0	0
Student 2	3	0.9839	0.0161	0
Student 3	5	0	1	1