Edge ML, also known as Edge Machine Learning, refers to the deployment of machine learning models on edge devices, such as smartphones, Internet of Things (IoT) devices, and other embedded systems, rather than relying on cloud-based servers for processing.

• In this project, I worked on providing solutions for TKElevator and Magna International using the proprietary MicroAI model which runs on edge devices like MCU/MPUs. I led the project in data sampling, sensor data analysis, and developed evaluation/business metrics for TKE and be in similar roles for Magna.
• Became a part of developing and deploying MicroAI Studio which enables users to fine-tune their MicroAI models on a web browser.

Anomaly or outlier detection is an important aspect of data science. Anomaly detection in sequential data has its own specific challenges due to seasonality and trends.

• Ensembled several statistical and machine learning models (Facebook Prophet, Seasonal and Trend Decomposition using Loess, Exponential Smoothing, ARIMA, SARIMA, Joint Estimation of Model Parameters and Outlier Effect)
• Developed heuristics to identify model parameters like seasonality.

Customer segmentation is the practice of partitioning the company's customers into discrete segments based on their certain traits, It is important for companies to have insights regarding their customers and it helps to identify potential customized marketing campaigns. Churn analysis is to understand why customers stop using your product and to predict before they churn.

• Research was granted by TUBITAK and published in 28th Signal Processing and Communications Applications Conference (SIU)
• Utilized common machine learning algorithms like Logistic Regression, XGBoost, LightGBM and Survival Analysis

Location analysis is a branch of operations research. Objective is to place available facilities in a way that minimize total transportation costs while considering constraints like avoiding placing hazardous materials near housing, and competitors' facilities. The techniques also apply to cluster analysis, coverage problem.

• Developed problem specific NSGA-2 algorithm at C++ to solve the location problem in a reasonable time. Github
• Visualizations and analysis are conducted at R.
• Linear math model is developed and solved using IBM CPLEX and C++.

Coverage Location Problems are often utilized when there is not enough budget, or human resources to cover all the demands. It seeks to cover a set of demands while one or more objectives are to be optimized.

• Developed genetic algorithm to solve well known Maximal Covering Location Problem. Report
• Compared proposed algorithm with math model and other meta-heuristic methods.

Payment Authorization is a way to ensure that the customer's account is valid. It is requested by the merchant to hold funds for the purchase. If authorization attempt is declined by the card issuing bank, and it is not possible to contact customer then the merchant may attempt another authorization. It is important for merchant to complete these recurring authorization attempts in minimum number which leads the cost and affect the customer experience. This analysis proposes a baseline model and approach to find better recycling practice.

• Three different approach is developed, Random Forest algorithm is used for generating empirical probability distributions. Blog + Codes
• Association Rule Mining, dfprophet, tsoutlier is utilized for Exploratory Data Analysis(EDA)

Presidential elections were held in Turkey in May 2023, alongside parliamentary elections, to elect a president for a term of five years. It marks the first time a Turkish presidential election has gone to a run-off. In this analysis, some questions regarding the opposition party leader's second round strategy are presented and answered using data of two consecutive election results.

• You can find the blog and codes here