This paper describes our winning solution for the ECML-PKDD ChAT Discovery Challenge 2020. We show that whether or not a Twitch user has subscribed to a channel can be well predicted by modeling user activity with boosting trees. We introduce the connection between target-encodings and boosting trees in the context of high cardinality categoricals and find that modeling user activity is more powerful then direct modeling of content when encoded properly and combined with a gradient boosting optimization approach.

Ensemble techniques--wherein a model is composed of multiple (possibly) weaker models--are prevalent nowadays within the field of machine learning (ML). Well-known methods such as bagging [1], boosting [2], and stacking [3] are ML mainstays, widely (and fruitfully) deployed on a daily basis. Generally speaking, there are two types of ensemble methods, the first generating models in sequence--e.g., AdaBoost [2]--the latter in a parallel manner--e.g., random forests [4] and evolutionary algorithms [5]. AdaBoost (Adaptive Boosting) is an ML meta-algorithm that is used in conjunction with other types of learning algorithms to improve performance. The output of so-called "weak learners" is combined into a weighted sum that represents the final output of the boosted classifier.

The book focuses on different variants of decision tree induction but also describes the meta-learning approach in general which is applicable to other types of machine learning algorithms. The book discusses different variants of decision tree induction and represents a useful source of information to readers wishing to review some of the techniques used in decision tree learning, as well as different ensemble methods that involve decision trees. It is shown that the knowledge of different components used within decision tree learning needs to be systematized to enable the system to generate and evaluate different variants of machine learning algorithms with the aim of identifying the top-most performers or potentially the best one. A unified view of decision tree learning enables to emulate different decision tree algorithms simply by setting certain parameters. As meta-learning requires running many different processes with the aim of obtaining performance results, a detailed description of the experimental methodology and evaluation framework is provided.

For supervised classification problems involving design, control, other practical purposes, users are not only interested in finding a highly accurate classifier, but they also demand that the obtained classifier be easily interpretable. While the definition of interpretability of a classifier can vary from case to case, here, by a humanly interpretable classifier we restrict it to be expressed in simplistic mathematical terms. As a novel approach, we represent a classifier as an assembly of simple mathematical rules using a non-linear decision tree (NLDT). Each conditional (non-terminal) node of the tree represents a non-linear mathematical rule (split-rule) involving features in order to partition the dataset in the given conditional node into two non-overlapping subsets. This partitioning is intended to minimize the impurity of the resulting child nodes. By restricting the structure of split-rule at each conditional node and depth of the decision tree, the interpretability of the classifier is assured. The non-linear split-rule at a given conditional node is obtained using an evolutionary bilevel optimization algorithm, in which while the upper-level focuses on arriving at an interpretable structure of the split-rule, the lower-level achieves the most appropriate weights (coefficients) of individual constituents of the rule to minimize the net impurity of two resulting child nodes. The performance of the proposed algorithm is demonstrated on a number of controlled test problems, existing benchmark problems, and industrial problems. Results on two to 500-feature problems are encouraging and open up further scopes of applying the proposed approach to more challenging and complex classification tasks.

Disclaimer: I'm a Senior Data Scientist at Saturn Cloud -- we make enterprise data science fast and easy with Python, Dask, and RAPIDS. Check out a video walkthrough here. Random forest is a machine learning algorithm trusted by many data scientists for its robustness, accuracy, and scalability. The algorithm trains many decision trees through bootstrap aggregation, then predictions are made from aggregating the outputs of the trees in the forest. Due to its ensemble nature, random forest is an algorithm that can be implemented in distributed computing settings.

Random forest (RF) is one of the most popular methods for estimating regression functions. The local nature of the RF algorithm, based on intra-node means and variances, is ideal when errors are i.i.d. For dependent error processes like time series and spatial settings where data in all the nodes will be correlated, operating locally ignores this dependence. Also, RF will involve resampling of correlated data, violating the principles of bootstrap. Theoretically, consistency of RF has been established for i.i.d. errors, but little is known about the case of dependent errors. We propose RF-GLS, a novel extension of RF for dependent error processes in the same way Generalized Least Squares (GLS) fundamentally extends Ordinary Least Squares (OLS) for linear models under dependence. The key to this extension is the equivalent representation of the local decision-making in a regression tree as a global OLS optimization which is then replaced with a GLS loss to create a GLS-style regression tree. This also synergistically addresses the resampling issue, as the use of GLS loss amounts to resampling uncorrelated contrasts (pre-whitened data) instead of the correlated data. For spatial settings, RF-GLS can be used in conjunction with Gaussian Process correlated errors to generate kriging predictions at new locations. RF becomes a special case of RF-GLS with an identity working covariance matrix. We establish consistency of RF-GLS under beta- (absolutely regular) mixing error processes and show that this general result subsumes important cases like autoregressive time series and spatial Matern Gaussian Processes. As a byproduct, we also establish consistency of RF for beta-mixing processes, which to our knowledge, is the first such result for RF under dependence. We empirically demonstrate the improvement achieved by RF-GLS over RF for both estimation and prediction under dependence.

As machine learning is increasingly used to help make decisions, there is a demand for these decisions to be explainable. Arguably, the most explainable machine learning models use decision rules. This paper focuses on decision sets, a type of model with unordered rules, which explains each prediction with a single rule. In order to be easy for humans to understand, these rules must be concise. Earlier work on generating optimal decision sets first minimizes the number of rules, and then minimizes the number of literals, but the resulting rules can often be very large. Here we consider a better measure, namely the total size of the decision set in terms of literals. So we are not driven to a small set of rules which require a large number of literals. We provide the first approach to determine minimum-size decision sets that achieve minimum empirical risk and then investigate sparse alternatives where we trade accuracy for size. By finding optimal solutions we show we can build decision set classifiers that are almost as accurate as the best heuristic methods, but far more concise, and hence more explainable.

This article provides an overview of Supervised Machine Learning (SML) with a focus on applications to banking. The SML techniques covered include Bagging (Random Forest or RF), Boosting (Gradient Boosting Machine or GBM) and Neural Networks (NNs). We begin with an introduction to ML tasks and techniques. This is followed by a description of: i) tree-based ensemble algorithms including Bagging with RF and Boosting with GBMs, ii) Feedforward NNs, iii) a discussion of hyper-parameter optimization techniques, and iv) machine learning interpretability. The paper concludes with a comparison of the features of different ML algorithms. Examples taken from credit risk modeling in banking are used throughout the paper to illustrate the techniques and interpret the results of the algorithms.

We used machine-learning classification models (classification decision tree and random forest model) and applied to a household survey. This was …

Unsatisfying accuracy of learning methods is mostly caused by omitting the influence of important parameters such as membership assignments, type of data objects, and distance or similarity functions. The proposed method, called Bounded Fuzzy Possibilistic Method (BFPM) addresses different issues that previous clustering or classification methods have not sufficiently considered in their membership assignments. In fuzzy methods, the object's memberships should sum to 1. Hence, any data object may obtain full membership in at most one cluster or class. Possibilistic methods relax this condition, but the method can be satisfied with the results even if just an arbitrary object obtains the membership from just one cluster, which prevents the objects' movement analysis. Whereas, BFPM differs from previous fuzzy and possibilistic approaches by removing these restrictions. Furthermore, BFPM provides the flexible search space for objects' movement analysis. Data objects are also considered as fundamental keys in learning methods, and knowing the exact type of objects results in providing a suitable environment for learning algorithms. The Thesis introduces a new type of object, called critical, as well as categorizing data objects into two different categories: structural-based and behavioural-based. Critical objects are considered as causes of miss-classification and miss-assignment in learning procedures. The Thesis also proposes new methodologies to study the behaviour of critical objects with the aim of evaluating objects' movements (mutation) from one cluster or class to another. The Thesis also introduces a new type of feature, called dominant, that is considered as one of the causes of miss-classification and miss-assignments. Then the Thesis proposes new sets of similarity functions, called Weighted Feature Distance (WFD) and Prioritized Weighted Feature Distance (PWFD).