🚀 Beginner ML Projects
A collection of beginner-friendly Machine Learning and Deep Learning projects I built while learning core ML concepts, algorithms, and deployment workflows. Each project includes data preprocessing, model training, evaluation, and in some cases, deployment.
📌 Project 1 — Student Placement Prediction (Logistic Regression)
Goal: Predict whether a student gets placed based on CGPA and IQ. Techniques Used:
Data preprocessing
Train–test split
Feature scaling
Logistic Regression
Decision boundary visualization
Model saving
Deployment using Streamlit
📌 Project 2 — Test Preparation Course Completion Prediction
Dataset: StudentsPerformance.csv (Kaggle) Features: Math score, Reading score Target: Test preparation course
0 → Not completed
1 → Completed
Techniques Used:
Logistic Regression
StandardScaler
Train/Test split
Decision boundary visualization
Model evaluation
Result: Achieved ~63% accuracy using only two numerical features.
📌 Project 3 — Spam Message Classification (Naive Bayes)
Built a text classification model using Naive Bayes to classify SMS messages as spam or ham. Dataset: Adapted from Kaggle SMS Spam Collection. Pipeline includes:
Text preprocessing (cleaning, tokenization, stopwords removal)
TF-IDF vectorization
Naive Bayes training
Evaluation metrics
📌 Project 4 — Breast Cancer Classification (Neural Network)
A binary classifier built using a simple Artificial Neural Network (ANN) to detect whether breast cancer is benign or malignant. Includes:
Feature scaling
Model building using deep learning
Train/validation split
Accuracy evaluation
📌 Project 5 — Face Mask Detection (Deep Learning)
Built a deep learning image classifier to detect the presence or absence of a face mask. Includes:
Image preprocessing
CNN model development
Data augmentation
Training & validation
Performance evaluation
📌 Project 6 — PINN Ball Trajectory Prediction (Physics-Informed Neural Network)
Goal: Predict the trajectory of a falling ball using a Physics-Informed Neural Network (PINN) that incorporates both data and the underlying physics (gravity) into training.
Techniques Used:
Data preprocessing and noisy observation handling
Neural network model building with input t and output h(t)
Physics-informed loss function incorporating the ODE: h''(t) = -g
Training using PyTorch with automatic differentiation
Model evaluation by comparing predicted trajectory to the exact solution
Visualization of results showing noisy data, exact solution, and PINN predictions
Result: Successfully learned a parabolic trajectory consistent with the physics of free fall, demonstrating the capability of PINNs to combine data and physical laws in predictive modeling.