Emre Akanak - Research

Long term goal of my research is to develop efficient models to predict social events; including customer behaviors, successful recommender systems, predicting stock exchange (behavior and prices), direction and range of commodity prices, predicting macroeconomic parameters, especially inflation and unemployment, budget deficits, BoT (Balance of Trade), etc. Crime rates and even specific predictive modeling on individuals and groups' behaviors.

Since the early years of university education, my main goal has been to understand the paths of human behaviors, including individuals, households, groups, social groups, cultural groups, political groups, and even nations.

During my second semester of economics education, I decided to dive deep into different social sciences and combine quantitative methodologies in economics. As an economist, I have studied in-depth to understand the methods and perspectives of other disciplines, including history, sociology, anthropology, and archeology, and combined multidimensional and interdisciplinary studies with econometrics, statistics, and mathematics.

Since the beginning of my first master's degree, I have preferred to focus on real-life case studies and use real-life empirical data during my academic research. During the first year of my Ph.D. education decided to study cliometrics in deep; since then, programming and AI have become part of my professional career in both academy and business.

My main goal is to advance in data science and combine my economics and econometrics studies with AI methodologies.

Modeling

Modeling includes many sub-studies and a deep understanding of quantitative methodologies and AI research. My focuses on modeling include developing unique and robust models, performance model stainless, setting customized parameters for model evaluation, and hyper fine-tuning hyperparameters.

Supervised Learning

Supervised Learning is one of my main priorities. Still, the DL structure is not very developed to filter external data and combine it with given data in Unsupervised Learning techniques. In case there is labeled data, Supervised Learning provides prediction abilities. The main key factor is strong EDA and modeling ability for highly accurate predictions.

Unsupervised Learning

Unsupervised Learning is applied methodologies for unlabeled data. My main observation is in unsupervised learning cases, EDA should be involved in a compelling and customized way.

Recommender Systems

Recommender Systems are being applied and used for creating targets from associated data. To develop an efficient recommendation system, it is essential to have an association and deep EDA knowledge.

Clustering

Clustering is one of the unsupervised learning methodologies which is mostly applied to understand group behaviors and customer segmentation studies. It is also being used for organized crime analysis, voter behavior analysis, and prediction. Multiclass clustering is the most advanced version of clustering studies, and it is fascinating to me.

Feature Engineering

Feature Engineering requires understanding the data and developing advanced transformation models. Also, it is highly related to vectorization but much more than it. It is possible to perform vectorization without feature engineering knowledge but not likely to apply feature engineering without vectorization. Especially in very complex cases, I use feature engineering.

Vectorization

Vectorization; is a kind of metric study to transform data to get better results. Effective and efficient vectorization requires high-level probability theory and Bayesian understanding.

ANN (Artificial Neural Networks) or (NNs)

ANN; is Artificial Neural Network (NNs), a Deep Learning methodology that mimics human learning mechanism. It is a supervised learning concept that requires network design. Unlike supervised learning, the system learns from the path of the data. I believe ANN studies need strong modeling knowledge, and I have been studying to understand different types of ANNs and develop unique and efficient networks. Especially PyTorch, TensorFlow, and Keras are very well-known libraries, and I make advances in related libraries.

Multiclass Clustering

Multiclass Clustering is like multiclass classification. Usually Scikit-Learn provides very well-established solutions for general cases. I have been working on understanding model structures. I also work on developing package for multiclass clustering and multiclass classification. Studies requires group theory and advanced distribution theory understanding.

Q2QT (Qualitative to Quantitative) Transformation

Q2QT (Qualitative to Quantitative) Transformation is one of the main and essential also very useful technique for machine learning. I have been studying on and advancing on it.