Online courses

Machine learning cannot do without algorithms and statistical methods. Aimed at the goal of predictive analytics, some of these algorithms and methods are core to machine or statistical learning. The course introduces these learning methods, covers their basics, components and variants, and shows how they can be practically applied in the context of regression and classification and how their results can be interpreted. Linear and logistic regression, regression splines for modeling non-linear relationships as well as regression-, similarity-, probability- and information-based classification are treated.

If you want to analyze data yourself, you need knowledge of data science concepts and methods of statistical learning on the one hand and skills in programming on the other. It is only when methodological and statistical core knowledge from data science meets practical programming skills that your own data analyses become possible. The course teaches the skills required for data analysis and, within this framework, sets the main topics in coordination with the participants. The range of topics includes working with R and R libraries (practical use of the infrastructure and programming options of R), data concepts and data management for preparing and performing data analyses (e.g. coding, recoding, subsetting, handling NAs, writing functions), for creating of graphics with R.

Two standards usually apply to data science analyses: they are primarily aimed at optimizing prediction accuracy and they place a strong emphasis on sound validation of the analysis. Accordingly, predictive analytics combined with sophisticated validation techniques are at the center of any data science analysis. In contrast, the goal of causal inference in connection with the management of data quality traditionally plays a major role in the methodology and research practice of empirical social science working with quantitative methods. The course addresses how, in the context of statistical learning, the goals of predictive analytics and causal inference can be combined and techniques of predictive, causal and statistical inference can be practically applied in a coherent manner.

Even if all the formal rules for drawing conclusions from data are followed, those conclusions can still be severely compromised by poor data quality. The reasons for this typically lie in selectively distorted selections, missing values and measurement errors. The course familiarizes with these sources of error, introduces the concepts and techniques used to solve them, and shows how these concepts and techniques can be practically applied using R and R libraries. It will be treated: Unit and item nonresponse and related missing data techniques (imputation methods and weighting procedures), random and systematic measurement error and related solutions (systematic replication, sensitivity analysis, latent variable analysis), management of selective data through the combination of probability and non-probability selections.