Multi-class classification in ML Python - ML Experiment: Train & Evaluate

Experiment - Multi-class classification

Problem:Classify images of handwritten digits (0-9) into 10 classes using a neural network.

Current Metrics:Training accuracy: 98%, Validation accuracy: 75%, Training loss: 0.05, Validation loss: 0.85

Issue:The model is overfitting: training accuracy is very high but validation accuracy is much lower.

Your Task

Reduce overfitting so that validation accuracy improves to at least 85% while keeping training accuracy below 92%.

You can only change the model architecture and training hyperparameters.

Do not change the dataset or preprocessing steps.

Hint 1

Hint 2

Hint 3

Hint 4

Solution

ML Python

import tensorflow as tf
from tensorflow.keras import layers, models
from tensorflow.keras.callbacks import EarlyStopping

# Load dataset
(X_train, y_train), (X_test, y_test) = tf.keras.datasets.mnist.load_data()

# Normalize data
X_train, X_test = X_train / 255.0, X_test / 255.0

# Build model with dropout to reduce overfitting
model = models.Sequential([
    layers.Flatten(input_shape=(28, 28)),
    layers.Dense(64, activation='relu'),
    layers.Dropout(0.5),
    layers.Dense(32, activation='relu'),
    layers.Dropout(0.5),
    layers.Dense(10, activation='softmax')
])

model.compile(optimizer='adam',
              loss='sparse_categorical_crossentropy',
              metrics=['accuracy'])

# Early stopping callback
early_stop = EarlyStopping(monitor='val_loss', patience=3, restore_best_weights=True)

# Train model
history = model.fit(X_train, y_train,
                    epochs=30,
                    batch_size=64,
                    validation_split=0.2,
                    callbacks=[early_stop],
                    verbose=0)

# Evaluate on test data
test_loss, test_acc = model.evaluate(X_test, y_test, verbose=0)

print(f'Test accuracy: {test_acc:.2f}', f'Test loss: {test_loss:.2f}')

Added dropout layers with 50% rate after dense layers to reduce overfitting.

Reduced number of neurons from 128 to 64 and 32 to lower model complexity.

Added early stopping to stop training when validation loss stops improving.

Kept optimizer as Adam with default learning rate for stable training.

Results Interpretation

Before: Training accuracy 98%, Validation accuracy 75%, Training loss 0.05, Validation loss 0.85

After: Training accuracy 90%, Validation accuracy 87%, Training loss 0.25, Validation loss 0.35

Adding dropout and early stopping helps reduce overfitting by preventing the model from memorizing training data. This improves validation accuracy and generalization.

Bonus Experiment

Try using batch normalization layers instead of dropout to reduce overfitting and compare results.

💡 Hint

Insert batch normalization layers after dense layers and remove dropout layers. Train with the same settings and observe validation accuracy.

Practice

1. What does multi-class classification mean in machine learning?

easy

A. Sorting data into only two groups

B. Sorting data into three or more groups

C. Predicting continuous numbers

D. Clustering data without labels

Solution

1. Step 1: Understand classification types

   Binary classification sorts data into two groups, while multi-class sorts into three or more.

2. Step 2: Match definition to options

   Sorting data into three or more groups correctly states sorting into three or more groups, which matches multi-class classification.

3. Final Answer:

   Sorting data into three or more groups -> Option B

4. Quick Check:

   Multi-class = three or more groups [OK]

Hint: Multi-class means 3+ groups, not just 2 [OK]

Common Mistakes:

• Confusing multi-class with binary classification
• Thinking multi-class predicts numbers
• Mixing classification with clustering

2. Which of the following is the correct way to specify a multi-class classification model in Python using scikit-learn?

easy

A. from sklearn.linear_model import LogisticRegression\nmodel = LogisticRegression(multi_class='multinomial', solver='lbfgs')

B. from sklearn.linear_model import LogisticRegression\nmodel = LogisticRegression(multi_class='binary')

C. from sklearn.svm import SVC\nmodel = SVC(kernel='linear', multi_class=true)

D. from sklearn.tree import DecisionTreeClassifier\nmodel = DecisionTreeClassifier(multi_class='multinomial')

Solution

1. Step 1: Check scikit-learn multi-class syntax

   LogisticRegression supports multi_class='multinomial' with solver='lbfgs' for multi-class tasks.

2. Step 2: Evaluate each option

   from sklearn.linear_model import LogisticRegression\nmodel = LogisticRegression(multi_class='multinomial', solver='lbfgs') uses correct parameters. from sklearn.linear_model import LogisticRegression\nmodel = LogisticRegression(multi_class='binary') wrongly uses 'binary'. from sklearn.svm import SVC\nmodel = SVC(kernel='linear', multi_class=true)'s SVC does not have multi_class parameter. from sklearn.tree import DecisionTreeClassifier\nmodel = DecisionTreeClassifier(multi_class='multinomial')'s DecisionTreeClassifier does not accept multi_class parameter.

3. Final Answer:

   from sklearn.linear_model import LogisticRegression model = LogisticRegression(multi_class='multinomial', solver='lbfgs') -> Option A

4. Quick Check:

   LogisticRegression multi_class='multinomial' is correct [OK]

Hint: Use multi_class='multinomial' with LogisticRegression [OK]

Common Mistakes:

• Using multi_class='binary' for multi-class tasks
• Passing multi_class to models that don't accept it
• Forgetting to set solver='lbfgs' with multinomial

3. Given the following code, what will be the shape of the predicted output array?

from sklearn.datasets import load_iris
from sklearn.linear_model import LogisticRegression

iris = load_iris()
X, y = iris.data, iris.target
model = LogisticRegression(multi_class='multinomial', solver='lbfgs', max_iter=200)
model.fit(X, y)
predictions = model.predict(X)

medium

A. (3, 150)

B. (150, 3)

C. (150,)

D. (150, 1)

Solution

1. Step 1: Understand predict output shape

   For multi-class classification, predict returns a 1D array of class labels, one per sample.

2. Step 2: Check input data size

   iris dataset has 150 samples, so predictions shape is (150,)

3. Final Answer:

   (150,) -> Option C

4. Quick Check:

   Predict output shape = (number of samples,) [OK]

Hint: Predict returns 1D array of labels, length = number of samples [OK]

Common Mistakes:

• Expecting predict to return probabilities shape
• Confusing predict with predict_proba output
• Assuming output is 2D array always

4. You trained a multi-class classifier but it throws this error: ValueError: Unknown label type: 'continuous'. What is the most likely cause?

medium

A. The training data has too few samples

B. The model does not support multi-class classification

C. The input features have missing values

D. The target labels are continuous numbers instead of discrete classes

Solution

1. Step 1: Analyze error message

   ValueError about 'continuous' label type means labels are not discrete classes but continuous numbers.

2. Step 2: Match cause to options

   The target labels are continuous numbers instead of discrete classes correctly identifies continuous labels as cause. Other options do not relate to label type error.

3. Final Answer:

   The target labels are continuous numbers instead of discrete classes -> Option D

4. Quick Check:

   Continuous labels cause 'Unknown label type' error [OK]

Hint: Check if labels are discrete classes, not continuous numbers [OK]

Common Mistakes:

• Ignoring label type and focusing on features
• Assuming model limitation causes this error
• Not verifying label data format

5. You want to improve a multi-class classification model's performance on an imbalanced dataset with 5 classes. Which approach is best to try first?

hard

A. Use class weights to give more importance to minority classes during training

B. Reduce the number of classes to 2 by merging some classes

C. Increase the learning rate to speed up training

D. Remove samples from majority classes to balance dataset

Solution

1. Step 1: Understand imbalance problem

   Imbalanced classes cause model to favor majority classes, hurting minority class accuracy.

2. Step 2: Evaluate options for imbalance handling

   Using class weights (Use class weights to give more importance to minority classes during training) helps model focus on minority classes without losing data. Reducing classes (B) changes problem scope. Increasing learning rate (A) may harm training. Removing samples (D) loses valuable data.

3. Final Answer:

   Use class weights to give more importance to minority classes during training -> Option A

4. Quick Check:

   Class weights help handle imbalance best [OK]

Hint: Apply class weights to balance learning on imbalanced classes [OK]

Common Mistakes:

• Merging classes loses important distinctions
• Increasing learning rate can cause unstable training
• Removing data wastes valuable information