Machine Learning

Artificial intelligence
• Intelligence displayed by machines that simulates animals & human
intelligence
• Data Science vs Machine Learning vs Artificial Intelligence
• Machine learning (Supervised, unsupervised & deep learning as well as
prediction). Subset of Artificial Intelligence
• Artificial intelligence (Automated decision making)
• Data science (Data transformation, Analytics insight & human decision
making). Use of statistical methods to find patterns in data
ML Topics
1. Intro to AI & Machine learning (Techniques & Applications)
2. Data wrangling & manipulation
3. Supervised learning – Regression & Classification
4. Feature engineering
5. Unsupervised learning
6. Time series modelling
7. Ensemble learning
8. Recommender systems
9. Text mining
Machine Learning techniques &
Applications
1. Classification • Healthcare (patient data & risk analytics,
medical imaging & diagnostics, wearables,
2. Categorization virtual assistants, in patient care, drug
discovery, lifestyle management & monitoring,
3. Clustering emergency room & surgery)
• Image processing (optical character processing
4. Trend analysis (Use time series – OCR, self driving cars)
data analysis) • Robotics (human simulation, humanoid robots,
industrial robotics)
5. Anomaly detection • Data mining (Anomaly detection, Grouping &
6. Visualization predictions, Association rules)
• Video games (reinforcement learning)
7. Decision making • Text analysis (sentiment analysis, spam filtering,
information extraction)
Data wrangling & manipulation (acquisition,
exploration, wrangling, manipulation & typecasting)
• Exploration techniques using Python – dimensionality check, dataset
types, slicing & indexing, unique elements, value extraction, feature
mean median and mode.
• Wrangling – techniques (discovering, structuring, cleaning,
enrichment, validation)
• Manipulation using python – coercion, merging, concantenation &
joining
Supervised learning – Regression &
Classification
• Classification – output has discreet and finite values
• Logistic regression/sigmoid probability (binary solutions – loan sanction, Pass/Fail, Customer segments,
Cancer prediction, Weather forecast). Accuracy metrics – Confusion matrix, ROC curve, AUC,
• Decision trees (sequential, hierarchical decisions). Root note, Decision node & Terminal node
• Random forest (ensemble of decision trees). Bagging – reduces variance & Bootstrapping. Produces
decorrelated decision trees
• Naïve Bayes classifier (Bayes theorem and assumption that features are independent)
• Support Vector Machines (hyperplane that separates instances into different categories)
• Regression – if output is continuous numeric variable such as R&D spend, Admin, Marketing
spend, State, Profit etc
• Linear
• Multiple linear
• Polynomial (Quadratic features)
• Ridge regression – when there is problem of collinearity
• Lasso regression – similar to ridge but also performs feature selection
• ElasticNet – combines Ridge & Lasso
Unsupervised learning
• Types of clustering (distance measures – complete, single, mean,
centroid)
• Hierarchical Clustering
• Agglomerative
• Divisive
• Partitioning Clustering
• K-Means (choose random k-data points as initial centroids, assign each datapoint closest
to centroid, calculate new cluster centroids, check if convergence criterion is met)
• Fuzzy C-Means
Feature engineering
• Factor analysis (extract underlying causes)
• PCA – Principal Component Analysis (reduces computational complexity).
Eigenvectors & Eigenvalues
• LDA – Linear Discriminant Analysis (reduces dimensions)
Time series modelling
• Time acts as an independent variable estimator
• Need – understand seasonal & cyclic patterns, detect unusual events,
forecast. Look out for white noise
• Stationarity (variance, covariance, mean). Use differencing & decomposition
to achieve stationarity
• Time series models
• Auto regressive models (AR) – predict using weighted sum of past values
• Auto regressive Integrated Moving Average (ARIMA) – linear combination of past and
present errors and values
• Moving Average (MA)
• Autocorrelation factors (ACF) & past autocorrelation factors (PACF)
Ensemble learning
• Combines individual models (averaging & weighted averaging – SVM,
Decision tree, Logistic regression)
• Bagging & bootstrapping aggregation algorithms (reduces variance by
taking mean of multiple estimates)
• Boosting – reduces bias by training weak learners sequentially
• Cross validation
• Adaboost
• Gradient boosting (GBM)
• Extreme Gradient Boosting (XGBoost)
Recommender systems (information
filtering technique)
• Collaborator filtering
• Item based nearest neighbor
• User based nearest neighbor
• Cosine & adjusted cosine similarity
• Association rule mining
• Apriori algorithm
Text Mining
• Text preprocessing – text transformation & attribute generation – attribute selection – data
mining or pattern discovery – interpretation or evaluation
• Applications
• Document clustering
• Products insight
• Pattern identification
• Security monitoring
• Natural language toolkit (NLTK) – tokenization, stemming, lemmatization, tagging, parsing &
semantic reasoning, named entity recognition (NER)
• N-grams (unigrams, bigrams & trigrams)
• NLP Process workflow
• Tokenization, stopword removal, stemming & lemmatization, POS tagging, information removal
• Structuring sentences
• Syntax, phase structure rules, Syntax tree parsing, rendering syntax tree, chunking & chunk parsing,
chinking (removing sequences of tokens from chunks
• Context free grammar (CFG)