🧠 Keras with Python Three-Layer Neural Network – Real-Life Student Result Prediction

🧠 Keras with Python

Three-Layer Neural Network – Real-Life Student Result Prediction

📘 What is Keras?

Keras is a high-level deep learning API used to build and train neural networks with Python. It provides easy-to-use tools for creating layers, selecting activation functions, training models and making predictions.

In this example, we will use Keras to create a neural network that learns from simple student information and predicts whether a student is likely to Pass or Need Improvement.

🎓 Real-Life Problem

Suppose a teacher wants to estimate a student's academic performance using three pieces of information:

• 📚 Study Hours per day
• 🏫 Attendance percentage
• 📝 Previous Marks percentage

The neural network will learn the relationship between these inputs and the student's result.

Study Hours + Attendance + Previous Marks
↓
Neural Network
↓
Pass / Need Improvement

🔷 Three-Layer Neural Network

The model contains three trainable layers:

📥

Layer 1
Input Layer
3 Features

➡️

⚙️

Layer 2
Hidden Layer
8 Neurons

➡️

📤

Layer 3
Output Layer
1 Neuron

Architecture:

3 Inputs → 8 Hidden Neurons → 1 Output

📊 Sample Training Dataset

Study Hours	Attendance %	Previous Marks %	Result
1	55	40	0
2	60	45	0
3	70	55	1
4	75	60	1
5	85	75	1

Here 0 = Need Improvement and 1 = Pass.

⚙️ Step 1: Install TensorFlow

Open the terminal or command prompt and run:

pip install tensorflow numpy

💻 Step 2: Complete Python Code

import numpy as np
from tensorflow import keras
from tensorflow.keras import layers

# --------------------------------
# 1. Training Data
# --------------------------------

X = np.array([
    [1, 55, 40],
    [2, 60, 45],
    [3, 70, 55],
    [4, 75, 60],
    [5, 85, 75],
    [6, 90, 80],
    [2, 65, 50],
    [4, 80, 70]
], dtype=float)

# 0 = Need Improvement
# 1 = Pass

y = np.array([
    0,
    0,
    1,
    1,
    1,
    1,
    0,
    1
], dtype=float)


# --------------------------------
# 2. Create Three-Layer Model
# --------------------------------

model = keras.Sequential([

    # Layer 1: Input + Hidden Layer
    layers.Dense(
        8,
        activation="relu",
        input_shape=(3,)
    ),

    # Layer 2: Second Hidden Layer
    layers.Dense(
        4,
        activation="relu"
    ),

    # Layer 3: Output Layer
    layers.Dense(
        1,
        activation="sigmoid"
    )
])


# --------------------------------
# 3. Compile the Model
# --------------------------------

model.compile(
    optimizer="adam",
    loss="binary_crossentropy",
    metrics=["accuracy"]
)


# --------------------------------
# 4. Train the Model
# --------------------------------

model.fit(
    X,
    y,
    epochs=300,
    verbose=0
)


# --------------------------------
# 5. Evaluate the Model
# --------------------------------

loss, accuracy = model.evaluate(
    X,
    y,
    verbose=0
)

print("Model Accuracy:", accuracy)


# --------------------------------
# 6. Test a New Student
# --------------------------------

new_student = np.array([
    [5, 88, 78]
], dtype=float)

prediction = model.predict(
    new_student,
    verbose=0
)

probability = prediction[0][0]

print("Pass Probability:", probability)

if probability >= 0.5:
    print("Prediction: PASS")
else:
    print("Prediction: NEED IMPROVEMENT")

⚠️ Understanding the Three Layers

In Keras, the model above contains three Dense layers:

1. First Dense Layer: 8 neurons with ReLU activation.
2. Second Dense Layer: 4 neurons with ReLU activation.
3. Third Dense Layer: 1 neuron with Sigmoid activation.

3 Inputs → 8 Neurons → 4 Neurons → 1 Output

Therefore, this example is a three-Dense-layer neural network.

🔍 Step-by-Step Code Explanation

1️⃣ Import Libraries

import numpy as np
from tensorflow import keras
from tensorflow.keras import layers

NumPy is used for numerical data, while Keras is used to construct and train the neural network.

2️⃣ Prepare Input Data

X = [
    [1,55,40],
    [2,60,45],
    [3,70,55]
]

Each row represents one student.

The three values represent:

• Study Hours
• Attendance
• Previous Marks

3️⃣ Create the First Layer

layers.Dense(8, activation="relu")

This layer contains 8 neurons. ReLU helps the network learn non-linear relationships.

4️⃣ Create the Second Layer

layers.Dense(4, activation="relu")

The second layer receives information from the previous layer and extracts more useful patterns.

5️⃣ Create the Output Layer

layers.Dense(1, activation="sigmoid")

The output layer contains one neuron. Sigmoid produces a value between 0 and 1, which can be interpreted as the estimated probability of the positive class.

⚡ Activation Functions

ReLU

ReLU(x) = max(0,x)

ReLU is used in the hidden layers to introduce non-linearity.

Sigmoid

σ(x) = 1 / (1 + e⁻ˣ)

Sigmoid converts the final output into a value between 0 and 1.

📐 Basic Mathematics

Every neuron first calculates a weighted sum:

z = w₁x₁ + w₂x₂ + w₃x₃ + b

Then the activation function is applied.

Hidden Layer → ReLU(z)

Output Layer → Sigmoid(z)

During training, Keras adjusts the weights and biases so that the predicted result becomes closer to the actual result.

🏋️ How Training Works

Student Data

➡️

Neural Network

➡️

Prediction

➡️

Loss

➡️

Weight Update

This process is repeated for many epochs. The optimizer Adam changes the model's parameters to reduce the loss.

🔮 Predicting a New Student

new_student = np.array([
    [5, 88, 78]
])

prediction = model.predict(new_student)

if prediction[0][0] >= 0.5:
    print("PASS")
else:
    print("NEED IMPROVEMENT")

The new student's information is:

• Study Hours = 5
• Attendance = 88%
• Previous Marks = 78%

The trained neural network calculates a probability. If the probability is at least 0.5, we classify the student as PASS.

🖥️ Example Output

Model Accuracy: 1.0

Pass Probability: 0.98

Prediction: PASS

Note: Neural-network results can vary slightly because the model starts with randomly initialized parameters and the dataset here is very small. This dataset is for educational demonstration, not for making real academic decisions.

📚 Important Keras Terms

Term	Meaning
Sequential	Creates a model where layers are arranged sequentially.
Dense	A fully connected neural-network layer.
ReLU	Activation function used in the hidden layers.
Sigmoid	Activation function that produces values between 0 and 1.
Epoch	One complete training pass through the dataset.
Optimizer	Algorithm used to update model parameters during training.

🎯 Learning Summary

1. Collect input features.
2. Prepare training data.
3. Create the Keras Sequential model.
4. Add the first Dense layer.
5. Add the second Dense layer.
6. Add the output Dense layer.
7. Compile the model.
8. Train the model using fit().
9. Evaluate the model using evaluate().
10. Predict the result of a new student using predict().

Real-Life Data → Keras Model → Learning → Prediction

🧠 Keras & Python Neural Network Learning Module
Created by Bijan Krishna Paul