Machine Learning Note:
Participants must complete at least
Tasks
3 tasks for the 2-week internship and
4 tasks for the 1-month internship
— from any level.
Level 1
Task 1: Student Score Prediction
Description:
Dataset (Recommended): Student Performance Factors (Kaggle
Build a model to predict students' exam scores based on their study hour
Perform data cleaning and basic visualization to understand the datase
Split the dataset into training and testing set
Train a linear regression model to estimate final score
Visualize predictions and evaluate model performance
Tools & Libraries:
Python Pandas Matplotlib Scikit-learn
Covered Topics
Regression | Evaluation metrics
Bonus:
Try polynomial regression and compare performance
Try experimenting with different feature combinations (e.g., removing or adding features like sleep, participation, etc.)
Task 2: Customer Segmentation
Description:
Dataset (Recommended): Mall Customer (Kaggle
Cluster customers into segments based on income and spending score
Perform scaling and visual exploration of groupings
Apply K-Means clustering and determine optimal number of clusters
Visualize clusters using 2D plots
Tools & Libraries:
Python Pandas Matplotlib Scikit-learn
Covered Topics
Clustering | Unsupervised learning
Bonus:
Try different clustering algorithms (e.g., DBSCAN)
Analyze average spending per cluster
1
Machine Learning
Tasks
Level 2
Task 3: Forest Cover Type Classification
Description:
Dataset (Recommended): Covertype (UCI
Predict the type of forest cover based on cartographic and environmental features
Clean and preprocess the data including categorical handling
Train and evaluate multi-class classification models
Visualize confusion matrix and feature importance
Tools & Libraries:
Python Pandas Scikit-learn XGBoost
Covered Topics
Multi-class classification | Tree-based modeling
Bonus:
Compare different models (e.g., Random Forest vs. XGBoost)
Perform hyperparameter tuning
Task 4: Loan Approval Prediction Description
Description:
Dataset (Recommended): Loan-Approval-Prediction-Dataset (Kaggle
Build a model to predict whether a loan application will be approved
Handle missing values and encode categorical features
Train a classification model and evaluate performance on imbalanced data
Focus on precision, recall, and F1-score
Tools & Libraries:
Python Pandas Scikit-learn
Covered Topics
Binary classification | Imbalanced data
Bonus:
Use SMOTE or other techniques to address class imbalance
Try logistic regression vs. decision tree
2
Machine Learning
Tasks
Level 2
Task 5: Movie Recommendation System Description
Description:
Dataset (Recommended): MovieLens 100K Dataset (Kaggle
Build a system that recommends movies based on user similarity
Use a user-item matrix to compute similarity scores
Recommend top-rated unseen movies for a given user
Evaluate performance using precision at K
Tools & Libraries:
Python Pandas Numpy Scikit-learn
Covered Topics
Recommendation systems | Similarity-based modeling
Bonus:
Implement item-based collaborative filtering
Try matrix factorization (SVD)
3
Machine Learning
Tasks
Level 3
Task 6: Music Genre Classification Description
Description:
Dataset (Recommended): GTZAN (Kaggle
Classify songs into genres based on extracted audio features
Preprocess features such as MFCCs or use spectrogram images
Train and evaluate a multi-class model using tabular or image data
If image-based, use a CNN model
Tools & Libraries:
Python Librosa (for features) Scikit-learn or Keras
Covered Topics
Audio data / CNNs | Multi-class classification
Bonus:
Try both tabular and image-based approaches and compare results
Use transfer learning on spectrograms
Task 7: Sales Forecasting Description
Description:
Dataset (Recommended): Walmart Sales Forecast (Kaggle
Predict future sales based on historical sales data
Create time-based features (day, month, lag values)
Apply regression models to forecast next period's sales
Plot actual vs. predicted values over time
Tools & Libraries:
Python Pandas Matplotlib Scikit-learn
Covered Topics
Time series forecasting | Regression
Bonus:
Use rolling averages and seasonal decomposition
Apply XGBoost or LightGBM with time-aware validation
4
Machine Learning
Tasks
Industry Level
Task 8: Traffic Sign Recognition Description
Description:
Dataset (Recommended): GTSRB (Kaggle
Classify traffic signs based on their image using deep learning
Preprocess images (resizing, normalization)
Train a CNN model to recognize different traffic sign classes
Evaluate performance using accuracy and confusion matrix
Tools & Libraries:
Python Keras TensorFlow
OpenCV
Covered Topics
Computer vision (CNN) | Multi-class classification
Bonus:
Add data augmentation to improve performance
Compare custom CNN vs. pre-trained model (e.g., MobileNet)
