JNTU B.Tech IV PREDICTIVE ANALYTICS Previous Question Paper (September - 2020 - R15)

The data analytics process is a systematic approach to transforming raw data into actionable insights. It involves the following key steps:

1. Problem Definition: Clearly define the business objective or the specific question to be answered. This ensures the analysis remains focused.
2. Data Collection: Gather relevant data from various sources such as databases, APIs, web scraping, or surveys.
3. Data Cleaning and Preparation: This is the most time-consuming step, involving the removal of duplicates, handling missing values, correcting inconsistencies, and formatting data for analysis.
4. Exploratory Data Analysis (EDA): Use statistical techniques and visualization tools to identify patterns, trends, outliers, and relationships within the data.
5. Data Modeling: Apply statistical or machine learning algorithms to build models that predict outcomes or classify data based on the identified patterns.
6. Interpretation and Communication: Translate technical findings into business-friendly insights, often using dashboards or reports, to support decision-making.
7. Deployment and Monitoring: Implement the model in a production environment and continuously monitor its performance to ensure accuracy over time.

Example: Customer Churn Prediction for a Telecom Company

A telecom provider wants to reduce customer attrition. The Problem Definition is to identify customers likely to leave. They perform Data Collection by pulling billing history, call logs, and customer support interactions. During Data Cleaning, they remove incomplete records and standardize usage metrics. In EDA, they discover that customers with high support ticket volumes are more likely to churn. They then build a Data Model using a classification algorithm (like Logistic Regression or Random Forest) to predict churn probability. Finally, they Communicate these insights to the marketing team, who then launch targeted retention offers to high-risk customers, and Monitor the churn rate to validate the model's effectiveness.

To construct the decision tree, we calculate the Information Gain for each attribute to determine the root node. The total dataset has 13 instances: 9 'yes' and 4 'no'. The entropy of the dataset is calculated as:

Calculating Information Gain for attributes:

1. Age: Entropy for groups (<=30: 2/5, 3/5; 31-40: 0/3, 3/3; >40: 2/5, 3/5). Gain(Age) ≈ 0.246.
2. Income: Entropy for groups (high: 2/4, 2/4; medium: 3/5, 2/5; low: 4/4, 1/4). Gain(Income) ≈ 0.029.
3. Student: Entropy for groups (No: 4/8, 4/8; Yes: 5/5, 0/5). Gain(Student) ≈ 0.151.
4. Credit_rating: Entropy for groups (fair: 6/8, 2/8; excellent: 3/5, 2/5). Gain(Credit_rating) ≈ 0.048.

Since Age provides the highest Information Gain, it is selected as the root node. The tree branches into Age groups: <=30, 31-40, and >40. For the <=30 branch, further splitting is performed using the 'Student' attribute. For the >40 branch, 'Credit_rating' is used to resolve the remaining impurity.

Overfitting Analysis: Overfitting occurs when a model captures noise or specific details of the training data rather than the underlying pattern, resulting in high training accuracy but poor generalization to new data. In this decision tree, if the tree is grown until every leaf node is pure (zero entropy), it will likely overfit. This is evidenced by the tree creating highly specific rules for small subsets of data. To mitigate this, techniques such as pre-pruning (setting a maximum depth or minimum instances per leaf) or post-pruning (removing branches that provide little predictive power) should be applied to ensure the model remains robust.