Search Results

You are looking at 1 - 3 of 3 items for :

  • "biased label problem" x
Clear All
Mr. Jorge A Chan-Lau

Front Matter Page Statistics Department Contents 1. Introduction 2. Machine Learning-Based Crisis Prediction Models 3. unFEAR: Unsupervised Feature Extraction Clustering for Crisis Regimes Classification 4. Application: Identification of economic crisis clusters 5. Conclusions References Figures 1. The biased label problem 2. Raw data clusters: non-separability and time clustering 3. Raw data clusters: country bias and imbalanced data 4. Two neural networks 5. The analogy between principal componetns analysis and the

Mr. Jorge A Chan-Lau

clustering method aimed at facilitating economic crisis prediction. The approach in unFEAR is quite different from that in other machine learning-based crisis prediction models. The latter adopt a supervised learning framework: at any time period, the models assign a crisis or no crisis label to a country’s observed economic and financial predictor data based on whether the observation was followed or not by a crisis n periods ahead. The reliance on labeled data gives rise to the biased label problem. Briefly, two countries characterized by similar economic and

Mr. Jorge A Chan-Lau
We introduce unFEAR, Unsupervised Feature Extraction Clustering, to identify economic crisis regimes. Given labeled crisis and non-crisis episodes and the corresponding features values, unFEAR uses unsupervised representation learning and a novel mode contrastive autoencoder to group episodes into time-invariant non-overlapping clusters, each of which could be identified with a different regime. The likelihood that a country may experience an econmic crisis could be set equal to its cluster crisis frequency. Moreover, unFEAR could serve as a first step towards developing cluster-specific crisis prediction models tailored to each crisis regime.