Recent years have witnessed an unprecedented increase in the availability of information on social, economic and health-related phenomena. Today researchers, professionals and policy makers have therefore, access to enormous databases (so-called Big Data) containing detailed information on individuals, companies and institutions and use of mobile devices. Machine learning is a relatively new approach to data analytics, which lies at the intersection between statistics, computer science and artificial intelligence. Its primary objective is that of turning information into knowledge and value by “letting the data speak”. In contrast to the more tradition approach of data analysis focusing on prior assumptions relating to data structure and the derivation of analytical solutions, Machine Learning techniques rely instead on a model-free philosophy development of algorithms, computational procedures, and graphical inspection of the data in order to more accurately predict outcomes. Computationally infeasible until very recently, Machine Learning is itself a product of the latest advancements in IT technology, of the computing power and the learning capabilities of today’s computers, of hardware development, and continuous software development.
This intensive introductory course offers an introduction to the standard machine learning algorithms currently applied to social, economic and public health data in order to illustrate (using a series of both official and user written Stata commands), how Machine Learning techniques can be applied to search for patterns in large (often extremely “noisy”) databases, which can subsequently be used to make both decisions and predictions.
In common with TStat’s training philosophy, each individual session is composed of both a theoretical component (in which the techniques and underlying principles behind them are explained), and an extensive applied (hands-on) segment, during which participants have the opportunity to implement the techniques using real data under the watchful eye of the course tutor. Throughout the course, theoretical sessions are reinforced by case study examples, in which the course tutor discusses and highlights potential pitfalls and the advantages of individual techniques. The intuition behind the choice and implementation of a specific technique is of the utmost importance. In this manner, the course leader is able to bridge the “often difficult” gap between abstract theoretical methodologies, and the practical issues one encounters when dealing with real data.
At the end of the course, participants are expected to be able to: i) autonomously implement (with the help of the Stata routine templates specifically developed for the course) the appropriate Machine Learning algorithms, given both the nature of their data and the analysis in hand, and ii) to have mastered the concepts of: factor-importance detection, signal-from-noise extraction, correct model specification and model-free classification, from both a data-mining an causal perspective.
Researchers and professionals working in biostatistics, economics, epidemiology, social and political sciences and public health wishing to implement Machine Learning techniques in Stata.
Participants should be familiar with the statistical software Stata. An introductory knowledge of econometrics and/or statistics is also required.
SESSION I: THE BASICS OF MACHINE LEARNING
Machine Learning: definition, rational, usefulness
Supervised vs. unsupervised learning
Regression vs. classification problems
Inference vs. prediction
Sampling vs. specification error
Coping with the fundamental non-identifiability of E(y|x)
Parametric vs. non-parametric models
The trade-off between prediction accuracy and model interpretability
Measuring the quality of fit: in-sample vs. out-of-sample prediction power
The bias-variance trade-off and the Mean Square Error (MSE) minimization
Training vs. test mean square error
The information criteria approach
Machine Learning and Artificial Intelligence
Integrating Python in Stata: an overview
SESSION II: RESAMPLING AND VALIDATION METHODS
Estimating training and test error
The validation set approach
Training and test mean square error
The Bootstrap algorithm
Bootstrap vs. cross-validation for validation purposes
SESSION III: MODEL SELECTION AND REGULARIZATION
Model selection as a correct specification procedure
The information criteria approach
Best subset selection
Backward stepwise selection
Forward stepwise Selection
Lasso and Ridge, and Elastic regression
Information criteria and cross validation for Lasso
SESSION IV: DISCRIMINANT ANALYSIS AND NEAREST-NEIGHBOR CLASSIFICATION
The classification setting
Bayes optimal classifier and decision boundary
Misclassification error rate
Linear and quadratic discriminant analysis
Naive Bayes classifier
The K-nearest neighbors classifier
SESSION V: NONPARAMETRIC REGRESSION
Beyond parametric models: an overview
Local, semi-global, and global approaches
Polynomial and series estimators
Partially linear models
Generalized additive models
SESSION VI:TREE-BASED REGRESSION
Regression and classification trees
Growing a tree via recursive binary splitting
Optimal tree pruning via cross-validation
Tree-based ensemble methods
Bagging, Random Forests, and Boosting
SESSION VII: NEURAL NETWORKS
The neural network model
neurons, hidden layers, and multi-outcomes
Training a neural networks
Back-propagation via gradient descent
Fitting with high dimensional data
Cross-validating neural network hyperparameters
Due to the current Public Health situation in Europe, we unfortunately have to reschedule this course date. We will be monitoring the virus situation very carefully over the forthcoming weeks, so as to be in a position to publish a feasible updated course schedule as soon as possible. Please accept our apologies for any inconvenience caused.
This intensive introductory course offers therefore an introduction to the standard machine learning algorithms currently applied to social, economic and public health data in order to illustrate (using a series of both official and user written Stata commands), how Machine Learning techniques can be applied to search for patterns in large (often extremely “noisy”) databases, which can subsequently be used to make both decisions and predictions.