Course 2 — Advanced AI & Machine Learning (60-Class Career Track)
Goal: Take students from ML fundamentals to model development, evaluation, optimization, deep learning and modern AI architectures. Prerequisite: Course 1 (AI/ML Foundation) or equivalent Python/Math assessment.
Core Engineering Competencies
Detailed Class-by-Class Syllabus
What is ML, ML lifecycle, Supervised, Unsupervised, Reinforcement learning, features, labels, train/val/test splits, overfitting, bias vs variance, complete ML workflow.
Traditional programming vs Machine Learning, when to use ML, industry use cases.
Problem definition, data acquisition, preprocessing, feature engineering, model training, evaluation, deployment.
Input features X, target vector y, labeled data, regression vs classification paradigms.
Pattern discovery in unlabeled data, clustering, dimensionality reduction intuition.
Agents, environment, states, actions, rewards, policy optimization overview.
Feature representation, tabular data, categorical vs numerical features, target variables.
train_test_split, preventing data leakage, validation set purpose.
High variance vs high bias, detecting overfitting through learning curves.
Bias-Variance tradeoff, model complexity vs generalization error.
Hands-on implementation of the complete ML workflow on a baseline dataset.
Project: Complete Scikit-Learn ML Workflow Pipeline
Dataset → Preprocessing → Feature Engineering → Train → Validation → Test → Evaluation → Model Export.
Regression fundamentals, Simple Linear Regression, Multiple Linear Regression, Polynomial Regression, cost function, MAE, MSE, RMSE, R² score, and House Price Prediction project.
Continuous target variables, predicting numerical quantities, regression line intuition.
y = mx + c, Ordinary Least Squares (OLS), calculating slope and intercept in Python.
Multiple independent variables, coefficients, multicollinearity, feature impact.
Modeling non-linear relationships, polynomial feature transformation, overfitting risks.
Implementing LinearRegression in Scikit-learn, checking assumptions of linear models.
Mean Absolute Error vs Mean Squared Error, penalizing large outlier errors.
Root Mean Squared Error interpretation, R-Squared & Adjusted R-Squared goodness-of-fit.
Real Estate & House Price Prediction: Cleaning → Feature Engineering → Linear Regression → Evaluation → Price Prediction.
Project: House Price Prediction System
Real Estate Dataset → Cleaning → Feature Engineering → Multiple Linear Regression → RMSE / R² Evaluation → Live Price Predictor.
Classification fundamentals, Logistic Regression, KNN, Decision Trees, Random Forest, Naive Bayes, SVM, Confusion Matrix, Precision, Recall, F1, ROC/AUC, and Churn Prediction project.
Binary vs Multiclass vs Multilabel classification, decision boundaries.
Sigmoid activation function, log-odds probability, binary decision threshold.
Distance-based classification, choosing optimal K, feature scale sensitivity.
Information gain, Gini impurity, entropy, tree splitting, pruning to prevent overfitting.
Ensemble bagging, bootstrap aggregating, voting classifier, feature importance scores.
Bayes theorem, conditional independence assumption, text spam classification.
Hyperplanes, maximum margin classifier, linear vs RBF kernel trick.
True Positives (TP), True Negatives (TN), False Positives (FP), False Negatives (FN).
Precision vs Recall tradeoff, F1-Score harmonic mean, accuracy paradox in imbalanced data.
ROC curve, Area Under Curve (AUC), comparing performance of 7 classifiers on single benchmark.
Customer Churn Prediction / Loan Approval Prediction / Spam Detection system.
Project: Telecom Customer Churn & Loan Approval Predictor
Build and compare Logistic Regression, Random Forest, and SVM models. Evaluate using Confusion Matrix, ROC-AUC, and export trained model.
Unsupervised learning, clustering, K-Means, Hierarchical Clustering, cluster evaluation (Elbow, Silhouette), PCA dimensionality reduction, and Customer Segmentation.
Clustering vs Dimensionality reduction, discovering hidden structures in unlabeled data.
Centroid-based, density-based, and hierarchical clustering methodologies.
K-Means centroid initialization, WCSS (Within-Cluster Sum of Squares), convergence algorithm.
Agglomerative vs Divisive clustering, dendrograms, linkage criteria (ward, complete, average).
Elbow method for optimal K, Silhouette score analysis (-1 to +1).
Principal Component Analysis, eigenvalues & eigenvectors, dimensionality reduction while preserving variance.
Customer Segmentation: Customer Data → Preprocessing → K-Means → Clusters → Customer Persona Segments.
Project: Customer Segmentation & Behavioral Clustering
Customer Demographic & Spending Data → Preprocessing → K-Means Clustering → Silhouette Evaluation → Targeted Marketing Segments.
Feature engineering, feature selection, scaling, K-Fold cross-validation, hyperparameter tuning, Grid Search, Random Search, and model optimization project.
Creating interaction features, binning, datetime feature extraction, domain-specific feature design.
VarianceThreshold, SelectKBest, Recursive Feature Elimination (RFE), feature importance ranking.
StandardScaler (Z-score) vs MinMaxScaler (0-1 normalization), RobustScaler for outlier resilience.
K-Fold Cross-Validation, Stratified K-Fold for imbalanced data, cross_val_score.
Parameters (weights) vs Hyperparameters (learning rate, tree depth, n_estimators).
GridSearchCV (exhaustive search) vs RandomizedSearchCV (efficient probabilistic search).
Bad Model → Identify Problem → Feature Engineering → Scaling → Cross Validation → Tuning → Better Model.
Project: Model Performance Optimization & Pipeline Engineering
Take a baseline 70% accurate model, apply feature engineering, scaling, Stratified K-Fold CV, and GridSearchCV to boost performance above 90%.
Introduction to Deep Learning, Neural Networks, Perceptron, Activation Functions, Forward Propagation, Loss Functions, Backpropagation, Gradient Descent, PyTorch/TensorFlow, and NN Classifier.
Why Deep Learning? Biological vs Artificial neurons, deep vs shallow architectures.
Input layer, hidden layers, output layer, weights, biases, dense/fully-connected layers.
Single layer perceptron, linear separability, XOR problem, multi-layer perceptron (MLP).
Sigmoid, Tanh, ReLU, Leaky ReLU, Softmax for multiclass output.
Matrix multiplications, dot products, layer-by-layer signal transmission.
Binary Cross-Entropy (Log Loss), Categorical Cross-Entropy, Mean Squared Error in deep learning.
Chain rule of calculus, calculating error gradients with respect to weights and biases.
Batch GD vs Stochastic GD (SGD) vs Mini-batch GD, learning rate, Adam optimizer.
Tensors, GPU acceleration, building neural network architectures in code.
Build a basic neural-network classifier for multi-feature tabular / image digit prediction.
Project: Deep Neural Network Classifier in PyTorch/TensorFlow
Build, train, and evaluate a multi-layer deep neural network with ReLU activations, Adam optimizer, and cross-entropy loss.
CNN & Computer Vision, RNN & LSTM, NLP fundamentals, Transformers introduction, Generative AI & LLM concepts.
Convolutional layers, filters, kernels, pooling (Max/Average), image feature maps, image classification.
Sequential data, memory cells, vanishing gradients, Long Short-Term Memory (LSTM) for time-series and text.
Tokenization, stop words, stemming, lemmatization, TF-IDF, Word2Vec embeddings, sentiment analysis.
Self-attention mechanism, encoder-decoder architecture, why Transformers revolutionized modern AI.
Foundation models (GPT, LLaMA), prompt engineering, temperature, hallucinations, RAG overview.
Final Capstone Project Development: Problem definition, data pipeline, feature engineering, model selection, tuning, evaluation, and live presentation defense.
Students build Capstone Choice (Churn / House Price / Segmentation / Image Classifier / Sentiment Analysis) following the complete 12-step pipeline.
1-on-1 viva defense, production code review, metrics evaluation, and Course 2 Advanced AI/ML Certification.
Course 2 Capstone: Production AI/ML Industry System
Problem Definition → Data Collection → Data Cleaning → EDA → Feature Engineering → Model Selection → Training → Evaluation → Optimization → Prediction → Presentation.