Acropolis Institute of Technology and Research, Indore
Department of Information Technology/ CSE(DS)
Project Problem Statements/ Descriptions
1. Customer Churn Prediction in the Telecom Sector
A leading telecom company is witnessing a steady decline in its subscriber base. The marketing
and customer care teams are under pressure to identify why customers are leaving and which
ones are most likely to leave in the next quarter.
Business Problem:
Telecom operators struggle with retaining customers, especially in saturated markets with
minimal differentiation in service offerings. Churn not only affects revenue but also indicates
potential issues in customer service, pricing, or product-market fit. Understanding customer
churn drivers enables proactive retention strategies and better service personalization.
Business Requirements:
Build a system to predict churn probability based on historical customer behavior and service
usage. Identify the most influential features contributing to churn.
Provide visual analytics to help marketing and customer support teams monitor at-risk
segments.
Deliverables:
[Link] dataset with derived features (e.g., tenure groupings, usage clusters).
[Link] models (Logistic Regression, Random Forest, XGBoost).
[Link] importance ranking and SHAP-based explainability plots.
[Link] dashboard to visualize churn prediction outcomes.
[Link] with business recommendations on reducing churn.
2. Retail Inventory Optimization using Demand Forecasting
A retail chain managing over 300 outlets across India struggles with mismatched supply and
demand, leading to frequent overstock and stockout situations, especially during festive
seasons.
Business Problem:
Retailers face challenges in managing inventory due to fluctuating customer demand, seasonal
variations, and supply chain inefficiencies. Overstocking ties up capital and leads to wastage,
while understocking causes missed sales and dissatisfied customers. Predictive modeling can
significantly optimize stock levels and reduce operational costs.
Business Requirements:
Forecast future demand for key products. Optimize restocking strategy and reduce inventory
holding costs. Visualize sales trends and flag potential overstock/understock scenarios.
Deliverables:
[Link]-level demand forecasts using ARIMA, Prophet, or LSTM.
[Link] replenishment model integrating reorder point and lead time.
[Link] for inventory trend analysis and restock alerts.
[Link] quantifying improvement in stock turnover ratio.
3. Fake News Detection Using NLP
A social media platform faces criticism for the unchecked spread of false information. The
platform wants to implement an AI-based mechanism to filter and flag potentially fake news
before it trends.
Business Problem:
In the digital era, the proliferation of fake news has led to misinformation crises, affecting
public trust and decision-making. Identifying false information before it spreads is crucial for
media platforms and news consumers.
Business Requirements:
[Link] an NLP-based system to classify news articles as fake or real.
[Link] confidence scoring and interpretability for each classification.
[Link] tools to analyze commonly faked topics or linguistic patterns.
Deliverables:
[Link] and labeled dataset of news headlines/articles.
[Link] classification models (TF-IDF + SVM, BERT-based classifier).
[Link] metrics (F1-score, AUC).
[Link] news analysis dashboard.
[Link] brief on fake news characteristics.
4. Loan Default Risk Prediction for a FinTech Startup
A newly launched FinTech app offers instant small-ticket loans to self-employed individuals.
However, rising defaults threaten investor confidence and operational sustainability.
Business Problem:
Fintech lenders operate in high-risk environments with minimal collateral and limited credit
history. Default risk jeopardizes the financial sustainability of lending platforms. A robust risk
assessment model is crucial for approving only eligible candidates.
Business Requirements:
Develop a credit scoring model based on borrower [Link] borrowers into risk
segments for [Link] model interpretability for compliance and audit
purposes.
Deliverables:
[Link]-processed loan dataset with imputed missing values.
[Link] models (XGBoost, LightGBM).
[Link] metrics (ROC-AUC, KS-statistic).
[Link] borrower scoring tool.
[Link] report with explainability visuals.
5. Personalized Movie Recommendation System
A regional OTT platform wants to increase its user retention rate. Despite having a rich library
of regional films and shows, users often complain about irrelevant recommendations.
Business Problem:
Streaming platforms face user churn due to overwhelming content options and poor
personalization. Providing intelligent, data-driven recommendations increases engagement,
satisfaction, and retention.
Business Requirements:
Build a recommender system that adapts to user [Link] historical ratings, genre
preferences, and user-item [Link] flexibility to handle new users or items (cold-start
problem).
Deliverables:
[Link] factorization or deep learning-based recommendation engine.
[Link]: RMSE, precision\@k, recall\@k.
[Link]-start strategy using content-based methods.
[Link]-end interface simulating a user profile and recommendations.
[Link] analysis of multiple recommender algorithms.
6. Employee Attrition Prediction in a Multinational IT Firm
A large IT services company is facing increased attrition among mid-level managers and
developers. HR wants to proactively identify employees likely to leave and take preventive
action.
Business Problem:
Employee attrition leads to increased recruitment costs, loss of organizational knowledge, and
project delays. Predictive analytics can help HR teams identify patterns and take early action
to retain valuable talent.
Business Requirements:
Build a model to predict employee attrition [Link] the main drivers influencing attrition
(e.g., salary, satisfaction, role).Provide actionable insights for HR to improve retention.
Deliverables:
[Link] and structured HR dataset with feature engineering.
[Link] prediction models (Decision Trees, Random Forest, Neural Networks).
[Link] impact analysis and visual explanation tools.
[Link] dashboard for real-time attrition risk monitoring.
[Link] report with retention recommendations.
7. Air Quality Index (AQI) Forecasting for Urban Planning
A municipal corporation wants to anticipate pollution spikes in major areas of the city to
implement timely public health advisories and traffic control measures.
Business Problem:
Urban pollution poses severe health risks, especially during winter. Accurate forecasting of
AQI levels enables better policy intervention and public warnings. Data science models can
predict future AQI values based on environmental and meteorological data.
Business Requirements:
Forecast AQI levels for specific city regions using historical pollution data. Identify key
contributing factors like PM2.5, PM10, traffic, and weather conditions. Provide visual alerts
and spatial heatmaps for urban planners.
Deliverables:
[Link]-series models (Prophet, ARIMA, LSTM) to forecast AQI.
[Link] heatmap showing relationships among pollutants.
[Link] visualizations using Folium/Plotly maps.
[Link] alert dashboard with prediction intervals.
[Link] suggesting policy and mobility improvements.
8. Dynamic Pricing Strategy for E-Commerce Platform
Scenario:
An e-commerce company struggles to stay competitive during sale seasons. Static pricing
strategies often result in lost revenue opportunities or unsold inventory. Management wants to
implement a dynamic pricing system based on real-time factors.
Business Problem:
E-commerce platforms face margin pressure from both high competition and changing
consumer behaviour. Rigid pricing models lead to suboptimal sales outcomes. Dynamic
pricing, powered by data science, can optimize revenue and improve customer conversion by
adjusting prices in real-time.
Business Requirements:
Build a dynamic pricing engine considering demand, seasonality, competitor pricing, and
stock levels. Ensure that price changes align with customer price sensitivity. Visualize
historical pricing vs. revenue trends.
Deliverables:
1. Cleaned and merged datasets (sales, web analytics, competitor prices).
2. Price elasticity models and demand forecasting (XGBoost, Ridge Regression).
3. Real-time pricing API simulation.
4. Dashboards showing predicted vs. actual revenue.
5. Business strategy document for pricing during high and low demand periods.
9. Credit Card Fraud Detection Using Machine Learning
Scenario:
A financial institution reports an increasing number of fraudulent transactions in credit card
usage, especially during online purchases. They seek a data-driven solution to detect anomalies
in real-time.
Business Problem:
Fraudulent transactions damage both customer trust and institutional credibility. Traditional
rule-based fraud systems lag behind sophisticated fraud techniques. Machine learning offers
real-time, adaptive fraud detection capabilities.
Business Requirements:
Build a classification model to detect fraudulent transactions. Ensure low false positives to
avoid annoying genuine customers. Integrate the model with transaction logs for real-time
scoring.
Deliverables:
1. Cleaned transactional dataset with anomaly labelling.
2. Classification models (Isolation Forest, Autoencoders, Logistic Regression).
3. Evaluation metrics: Precision, Recall, Confusion Matrix.
4. Real-time anomaly detection dashboard.
5. Model deployment plan with alert mechanisms.
10. Sales Conversion Prediction for a B2B CRM Platform
Scenario:
A B2B SaaS company uses a CRM platform to manage sales leads. The sales team struggles
to prioritize leads based on conversion probability, leading to poor productivity and lower
conversion rates.
Business Problem:
Sales teams waste time chasing cold leads without insights into potential deal closure.
Predicting lead conversion probability helps in effective resource allocation and closing more
deals.
Business Requirements:
Analyze historical lead data to predict conversion likelihood. Identify key lead attributes
contributing to success. Provide visual tools for the sales team to score and rank leads.
Deliverables:
1. Cleaned CRM data with merged sales outcomes.
2. Predictive models (Logistic Regression, Gradient Boosting).
3. SHAP values to explain model predictions.
4. Sales prioritization dashboard with filters.
5. Report on lead nurturing strategies and model impact.
11. Predictive Maintenance for Manufacturing Equipment
Scenario:
A manufacturing plant incurs heavy costs due to sudden machinery breakdowns. Traditional
time-based maintenance is inefficient. They want to switch to predictive maintenance using
IoT sensor data.
Business Problem:
Unexpected equipment failures lead to operational downtime and losses. Predictive
maintenance enables preemptive interventions, reducing repair costs and improving efficiency.
Business Requirements:
Predict potential equipment failures based on sensor logs. Identify key patterns leading to
breakdowns. Provide real-time alerts to maintenance teams.
Deliverables:
1. Processed sensor data with timestamps and anomaly tagging.
2. Predictive models (Random Forest, LSTM).
3. Anomaly detection module with alert thresholds.
4. Maintenance recommendation dashboard.
5. Report on reduction in downtime and cost impact.
12. Customer Segmentation for a Retail Loyalty Program
Scenario:
A supermarket chain wants to launch a personalized loyalty program. However, it lacks proper
customer segmentation and spends uniformly on all customers, reducing campaign
effectiveness.
Business Problem:
Loyalty programs without targeted segmentation waste resources. Segmenting customers by
purchasing behaviour helps in designing effective, personalized campaigns to improve
retention and profitability.
Business Requirements:
Segment customers based on purchasing frequency, recency, and monetary value. Identify
high-value and dormant segments. Recommend tailored offers for each segment.
Deliverables:
1. RFM analysis and clustering models (KMeans, DBSCAN).
2. Customer segments labeled with characteristics.
3. Dashboard for segment-wise behavior tracking.
4. Loyalty strategy recommendation report.
5. Campaign simulation with A/B testing outcomes.
13. Social Media Sentiment Analysis for Product Launch
Scenario:
A consumer electronics company launches a new product. They want to analyze public
sentiment from Twitter and Instagram to assess brand perception and campaign effectiveness.
Business Problem:
Companies need to monitor online sentiment post-launch to evaluate marketing impact and
detect PR crises early. Manual tracking is inefficient for large-scale campaigns.
Business Requirements:
Collect and clean social media data based on keywords/hashtags. Analyze sentiment trends and
emerging topics. Provide visual analytics for the marketing team.
Deliverables:
1. Data collection using APIs (Twitter, Instagram).
2. Sentiment classification models (VADER, BERT).
3. Word clouds, sentiment timelines, and influencer impact visuals.
4. Real-time sentiment dashboard.
5. Brand perception and feedback analysis report.