ADVANCE AIML CIE -03 ANSWERS(21AI71)

MODULE-03
1.What is the basic structure of a neural network, and how are weights and activation functions used in its
representation?

ANS. Basic Structure of a Neural Network:

A neural network is made up of layers of nodes (called neurons). It has:

 Input Layer: This is where the data enters the network. Each input neuron represents one feature of
the data.
 Hidden Layers: These layers between the input and output layers help process the data and learn
patterns. There can be one or more hidden layers.
 Output Layer: This layer gives the final result or prediction, like a classification (e.g., "dog" or "cat").
 Weights are values that control how strongly the input data affects the output. When data moves
through the network, the weights are adjusted to help the network learn.
 Activation Functions are math formulas that decide if a neuron should "fire" (or be activated). They
help the network understand complex patterns. Common ones are ReLU and sigmoid.

2. Describe the key steps involved in the backpropagation algorithm for training multilayer neural networks.

ANS. Key Steps in Backpropagation:

Backpropagation is how a neural network learns by adjusting its weights to improve its predictions. Here's how
it works:

 Forward Pass:The input data moves through the network, and the network makes a prediction.
 Calculate Error:The network compares its prediction with the correct answer (target). The difference
between them is called the error.
 Backpropagation (Backward Pass):The error is sent backward through the network. The network
calculates how much each weight contributed to the error.
 Update Weights:The weights are adjusted based on the error. This helps the network make better
predictions next time.
 Repeat:Steps 1-4 are repeated many times (over many "epochs") until the network makes accurate
predictions.
 MODLE-04
1.What are association rules in data mining, and how are metrics like support, confidence, and lift used to
evaluate them?

ANS. What is an Association Rule?

An *association rule* is an "if-then" statement that describes relationships between items in a dataset, such as
"if a customer buys bread, then they are likely to buy butter." This analysis helps identify patterns in
purchasing behavior.

Three Key Metrics Used to Generate Association Rules

1.Support:

- Definition: Measures how frequently an itemset appears in the dataset.

- Example: If 100 transactions occur and 20 include both bread and butter, the support for "bread → butter"
is 20/100 = 0.2 (or 20%).

2. Confidence:

- Definition: Indicates how often the consequent occurs when the antecedent is present.

- Example: If 15 out of 20 transactions that include bread also include butter, the confidence for "bread →
butter" is 15/20 = 0.75 (or 75%).

3. LifT:

- Definition: Measures how much more likely the consequent occurs with the antecedent compared to when
it is not present.

- Example: If the overall probability of buying butter is 10%, and confidence for "bread → butter" is 75%,
then lift = 0.75 / 0.10 = 7.5. A lift greater than 1 indicates a strong association.

Q2. Differentiate between user-based and item-based collaborative filtering. What are the key advantages and
limitations of each?

ANS
3.How can the Surprise library be used for building collaborative filtering models? Provide a brief overview of its main
components and functionality

ANS. Using the Surprise Library for Collaborative Filtering

The Surprise library in Python helps you create recommendation systems. These systems suggest items (like movies or
products) based on what users like. Here's a simple breakdown of how it works:

Key Components of Surprise:

Dataset:

What it is: Surprise helps you load data of users rating items (like movies or products). You use this data to make
recommendations.

Example: A dataset might contain ratings like "User A rated Movie X 4 stars."

Algorithms:

What it is: Surprise provides ready-made algorithms to make recommendations based on user preferences.

Popular Algorithms:

 KNN (K-Nearest Neighbors): Finds similar users or items and suggests them.
 SVD (Singular Value Decomposition): A technique to predict ratings by breaking down data into smaller parts.
 Evaluation:
 What it is: After building the model, you can check how good it is by comparing predicted ratings with actual
ratings using metrics like RMSE (Root Mean Square Error).
 Example: It helps you know if the recommendations are accurate.

Training & Testing:

What it is: You train the model on some data and test it on new data to see how well it performs.

How to Use Surprise:

 Load Data: Import your rating data into Surprise (e.g., who liked which movie).
 Pick an Algorithm: Choose an algorithm (like KNN or SVD) to make recommendations.
 Train the Model: Use the data to train the model (teach it how to recommend items).
 Make Predictions: After training, the model can suggest items for users based on their past ratings.
 Evaluate the Model: Check how well the model made predictions using metrics like RMSE.

Key Features of Surprise:

 Easy to Load Data: Quickly load your data in the right format.
 Pre-Built Algorithms: Use algorithms like KNN or SVD for making recommendations.
 Evaluate Models: Test how accurate the model’s recommendations are.
 Customizable: You can tweak the algorithms to better suit your needs.
 MODULE -05
1.Explain the differences between hierarchical clustering and partitioning methods in terms of their
approaches to grouping data.

ANS.

2. How does the k-means algorithm work? Highlight its initialization, iteration, and stopping criteria.

ANS . How the K-Means Algorithm Works

K-Means is a popular clustering algorithm that groups data into a fixed number of clusters (denoted as k).
Here's how it works:

Initialization:

Choose k: Decide how many clusters (k) you want.

Randomly select k centroids (points that represent the center of each cluster). These can be randomly chosen
data points or just random positions.

Iteration:

Assign data points to clusters: For each data point, calculate the distance to each centroid and assign it to the
nearest centroid.

Recalculate centroids: After all points are assigned, update the centroids by calculating the average of all
points in each cluster. This becomes the new centroid for each cluster.

Stopping Criteria:

The algorithm stops when the centroids no longer change significantly between iterations (or after a set
number of iterations). This means the clusters are stable and the algorithm has converged.
Summary:

Initialization: Choose k centroids randomly.

Iteration: Assign points to the nearest centroid, then update centroids.

Stopping: When centroids stop changing or after a set number of iterations.

3.Describe the steps involved in agglomerative hierarchical clustering and how it differs from divisive
hierarchical clustering.

ANS. Steps in Agglomerative Hierarchical Clustering & Differences with Divisive Clustering

Agglomerative Hierarchical Clustering:

Start with individual points: Each data point starts as its own cluster.

Merge the closest clusters: Find the two clusters that are most similar (based on a distance measure like
Euclidean distance) and merge them into a single cluster.

Repeat: Keep merging the closest clusters step-by-step until all data points are in a single cluster.

Create a Dendrogram: This is a tree-like diagram showing how clusters were merged at each step.

Divisive Hierarchical Clustering (Opposite Approach):

Start with one big cluster: All data points start in a single cluster.

Split the cluster: Split the most dissimilar (or diverse) cluster into two smaller clusters.

Repeat: Continue splitting the largest cluster until each data point is in its own cluster.

Create a Dendrogram: Just like agglomerative, a tree shows how the data was split.

Key Differences:

 Agglomerative: Starts with individual points and merges clusters together.