Gokhan Ciflikli. Unpacking Classification Algorithms: Local and Global Explanations.
Machine learning approaches to conflict studies predominantly rely on global metrics such as predictive accuracy and variable importance to establish validity. However, previous research shows that they are prone to bias in the presence of varying scales and collinearity. I compare local and global explanations of black box classification algorithms and offer a set of guidelines for social scientists.
Gokhan Ciflikli and Nils Metternich. Taking time seriously when evaluating predictions in Binary-Time-Series-Cross-Section-Data. [Pre-Print]
We demonstrate that standard classification metrics for binary outcome data are prone to underestimate model performance in a binary-time-series-cross-section context. We argue for temporal residual based metrics to evaluate cross-validation efforts in binary-time-series-cross-section and test these in Monte Carlo experiments and existing empirical studies.
Nils Metternich, Gokhan Ciflikli, and Ali Altaf. Predicting the severity of civil wars: An actor-centric approach. [Pre-Print]
This paper introduces RebelTrack and RebelCast, which consist of a set of functions to measure rebel behaviour, characteristics, and configurations. Using mainly data from the UCDP GED dataset in combination with PRIO grid data, RebelTrack establishes an open-development platform to generate theory and data-driven information on rebel organizations. RebelCast is an intuitive random forest fitter for the resulting RebelTrack output using cross-validation techniques suitable for time-series-cross-section data.
Gokhan Ciflikli. 2018. Learning Conflict Duration: Insights from Predictive Modelling. PhD thesis, The London School of Economics and Political Science (LSE). [PDF]