Election Exit Poll and Voter Prediction Model

Problem Statement : You are hired by one of the leading news channels CNBE who wants to analyze recent elections. This survey was conducted on 1525 voters with 9 variables. You have to build a model, to predict which party a voter will vote for on the basis of the given information, to create an exit poll that will help in predicting overall win and seats covered by a particular party.

Data Dictionary for Dataset

Feature	Description	Interpretation
vote	The political party the voter supports. Values: "Labour" or "Conservative"	This is your target variable — what you're trying to predict using all other features.
age	Age of the voter in years.	A numeric feature. Older or younger voters may prefer different parties.
economic.cond.national	Rating of the country’s economic condition (1–5 scale).	Subjective opinion: 1 = very bad, 5 = very good. May influence party preference based on national performance.
economic.cond.household	Rating of personal/household financial situation (1–5 scale).	Also subjective: how well-off the respondent feels economically. Important for voter behavior.
Blair	How the respondent rates Tony Blair, leader of the Labour Party at the time (1–5 scale).	1 = dislike very much, 5 = like very much. A high score suggests voter favors Labour.
Hague	Rating of William Hague, leader of the Conservative Party (1–5 scale).	Similarly, higher scores suggest preference for the Conservatives.
Europe	11-point scale on attitude toward European integration. Higher = more Eurosceptic (less support for EU).	Political alignment with EU-related issues. Can influence support for parties based on EU stance.
political.knowledge	How well the respondent understands each party’s EU position (0–3 scale).	0 = no knowledge, 3 = high knowledge. May correlate with how confidently a person votes.
gender	Gender of the voter: "male" or "female".	Used to analyze demographic influence on vote preference.

💡 Example Interpretation: A person who:

Rates Blair = 5, Hague = 1

Thinks national economy = 4, household = 4

Has low Euroscepticism (Europe = 2) ... is likely to vote Labour, based on ideology and leadership preference.

Steps Performed

1. Data Loading and Initial Exploration
2. Dropping Irrelevant Column (Serial Number)
3. Checking for Duplicates and Removing Partial Duplicates
4. Exploratory Data Analysis (EDA): Generated Automated EDA Report using ydata-profiling
5. Outlier Detection via Boxplots
6. Encoding Categorical Variables (Vote and Gender)
7. Feature Scaling for Numerical Columns
8. Data Splitting (70 (Train) :30 (Test) with Stratification)
9. Model Building and Evaluation:
   • Logistic Regression
   • Linear Discriminant Analysis (LDA)
   • k-Nearest Neighbors (kNN)
   • Naive Bayes
   • Random Forest (Bagging)
   • General Bagging
   • Boosting (Adaptive Boosting)
10. Evaluation Metrics: Accuracy, Confusion Matrix, Classification Report
11. Final Model Selection and Conclusion

Vote Share and Exit Poll Insights

• Labour has strong support and Vote Share.

  

• Female voters are more active and politically engaged in this dataset.

  

• Age-wise visualization showed younger and middle-aged groups favor Labour more.

  

• These trends suggest the Labour party's campaign had wider appeal and engagement. Clear Win of Labour Party.

  

Exploratory Data Analysis (EDA)

See Election_EDA_Report.html for Automated EDA Report and EDA revealed the following insights:

✅ Labour Party :

• Both females and males voted predominantly for Labour.

• Females cast more votes than males for Labour.

• Labour shows stronger appeal across all genders, particularly female voters.

✅ Conservative Party :

• Overall, fewer votes were cast for Conservatives compared to Labour.

• Females again outvoted males for Conservatives, though the gap is smaller than in Labour.

✅ Overall Gender Voting Trend :

• Female voters are more active than male voters in this dataset (consistent with the previous bar chart: 808 females vs. 709 males).

• Female votes outnumber male votes across both political parties.

• Labour maintains broader gender support, while Conservative support is more limited and shows a lower male participation.

📊 Age Group-Based Voting Trends :

Younger Age Groups (18–30 and 31–45) :

• Show strong support for Labour, suggesting that younger voters lean more progressive or are more influenced by Labour’s messaging.

• Labour gains a significant portion of its votes from these age groups.

Middle Age Group (46–60) :

• Slightly more balanced between the two parties but still shows Labour in the lead.

• This group might reflect working-class concerns that resonate more with Labour policies.

Older Age Group (61+) :

• Conservative support increases slightly here, but Labour still maintains a lead.

• This could suggest some generational alignment with traditional conservative values but not enough to surpass Labour.

🎯 Conclusion :

The Labour Party dominates across both gender and age groups, making it the most popular party in this dataset and Clear Win of Labour Party.

• Female voters are the most engaged, casting more votes than males in both parties.

• Labour’s success is especially strong among young voters, while Conservatives perform relatively better among older demographics, but not enough to close the gap.

These trends could reflect differences in political priorities, outreach strategies, or social engagement patterns across demographics.

Model Comparison and Inference

Model	Training Accuracy	Test Accuracy	Notes
Logistic Regression	0.836	0.840	Solid baseline; balanced precision/recall.
Linear Discriminant Analysis (LDA)	0.838	0.842	Slightly better than logistic regression; well balanced.
k-Nearest Neighbors (kNN)	0.852	0.836	Highest training accuracy (risk of overfitting).
Naive Bayes	0.825	0.844	Lowest training accuracy but good generalization.
Random Forest (n=1000)	—	0.827	Stable and robust, but slightly underperforms others.
Bagging (n=1000)	—	0.827	Similar to Random Forest, consistent but not top-performing.
Adaptive Boosting (n=1000)	—	0.849	⭐ Best test accuracy; strong performance overall.

Among all models, Boosting delivered the highest test accuracy (84.9%), indicating that it is the best performing model for this dataset.

Inference

• Adaptive Boosting outperformed all other models in terms of test accuracy (84.9%), making it the best overall. It achieved a good balance of precision and recall.
• Naive Bayes surprised with a strong test performance despite a lower training score — indicating excellent generalization.
• LDA and Logistic Regression were also strong contenders with consistent and high accuracy, showing they are still competitive for linear separability.
• kNN had the highest training accuracy, suggesting it might be overfitting slightly — as its test accuracy dropped a bit.
• Random Forest and Bagging had solid but slightly lower performance, possibly due to parameter defaults or redundancy in tree-based ensembles for this dataset.

Conclusion

• Best Performing Model: Boosting — Adaptive Boosting — due to its superior accuracy and f1-score.
• Caution : Models like kNN and Random Forest can overfit if hyperparameters aren’t tuned.
• Recommended Final Model: Boosting, with potential tuning for learning_rate and n_estimators=1000.

Streamlit Deployment

A user-friendly Streamlit web application was created to allow users to input voter features and get real-time predictions using the trained AdaBoost model voter_model.pkl.