Brand loyalty is an important factor for companies to consider, as it can affect their ability to generate revenue and grow their business. The purpose of this study is twofold: first, it aims to examine the determinants of brand loyalty in the Fast-Moving Consumer Goods (FMCG) sector using household consumer panel data. To do this we propose a definition of brand loyalty based on household consumption volumes in different FMCG channels and then utilize machine learning techniques including Fuzzy C-Means (FCM) to segment customers based on their purchasing behavior, predict consumers' brand loyalty, and identify the factors influencing brand loyalty. Secondly, we aim to investigate the relationship between store loyalty and brand loyalty using Propensity Score Matching (PSM). Results reveal that customers who show loyalty to stores are also more likely to show loyalty to particular brands.

Electricity load forecasting has served as the foundation for predictive and prescriptive analytics problems in the energy analytics domain. Accurate forecasts of the electricity demand provide an important advantage in estimating the hourly market clearing price for electricity since it can be seen as the main driver for its fluctuations. Such forecasts can be inputs to many optimization problems related to portfolio optimization for a power producer. In this study, short term electricity demand will be taken into consideration as a multivariate series forecasting problem. Hourly electricity consumption data starting from January 2016 up to January 2025 from Turkey has been included in the experiments. Several deep learning algorithms such as Temporal Fusion Transformer, N-Beats and NHits has been used alongside a relatively more conventional forecasting approach, LightGBM. A model selection technique that is developed for High-Frequency Trading domain, Combinatorial Purged K-Fold Cross Validation will be extended into a problem with a non-financial dataset.

This study examines implementing a machine-learning-based system for customer segmentation in the apparel retail industry through the use of image-processing technologies powered by sensors located in retail stores. In this study, analyses have been conducted based on two primary raw data that includes reallife footfall and purchase data from a retail store over a three-month period in 2024. In this study, a deeper understanding of customer behavior has been gained upon conducting data preprocessing and feature engineering techniques on raw datasets including footfall data capturing the temporal locations of each unique visitor and transaction data. Model design and training were performed, applying five widely used machine learning algorithms for customer segmentation both in train and test sets. The process also included testing, validation, and cluster analysis to derive meaningful customer insights. Thus, this study introduces a novel perspective on analyzing in-store customer behavior and its shopping patterns by focusing on the temporal tracking of customers' locations within the store environment. Accordingly, the distinctive data preprocessing and feature engineering techniques enabled better understanding of customer behavior and let the decision-makers offer customers a more tailored journey while maximizing customer satisfaction, engagement, and business profitability.

The pursuit of understanding the science behind the success of a song has been a challenge for decades. Hit Song Prediction (HSP), a subfield of Music Information Retrieval, helps artists, labels, and talent scouts predict song performance and streamline market-driven song selection. The purpose of this study is to suggest a new prediction model which has the ability to detect the top 10 songs out of Billboard Hot 100 songs, using a multi-model approach. Using a dataset of 300 charted songs of last 2 years, we have developed a range of ML models including Gradient Boosting, Multi-Layer Perceptron (MLP) and Decision Trees. Analysis incorporated lyrics, audio characteristics, and artist-related data including social media metrics. Results reveal that combining audio, lyrics and social media data is a promising strategy in HSP.

Over the years, similar to global markets, the Turkish electricity market has undergone significant reforms in order to meet the need for a more liberal market. One of these changes has been the establishment of a day-ahead market which is an organized market operated by the Market Operator, facilitates electricity trading and balancing activities one day prior to the delivery of electricity. In this market, the price that emerges in the hourly segments where the supply and demand curves intersect is called the Market Clearing Price. The growing need for optimization in areas such as more effective bid strategies, risk management, and increasing competition necessitated more accurate predictions of the MCP. The purpose of this study is to suggest market clearing price prediction models which are sensitive to price fluctuations using machine learning techniques. The dataset, which contains Market Clearing Price data, covers the period from January 2020 to December 2024. Furthermore, using feature engineering, external factors like lagged and nonlagged features, temporal indicators (weekday, weekend, and holiday), and renewable energy output will be included. The performance of the models is evaluated using metrics like R2, Mean Absolute Error, and Root Mean Squared Error.

Demand forecasting is a vital problem that affects nearly every aspect of supply chain management operations in the retail industry. A retailer would need to precisely anticipate the upcoming demand to efficiently manage their inventory, form profitable pricing strategies, and handle logistics operations in time. Novel techniques in machine learning and deep learning literature have also been important in demand forecasting domain as they thrive at extracting complex relationships from data. In this study, a multivariate hierarchical time series forecasting problem is tackled for a leading retailer that operates with hundreds of stores in Turkey. A dataset consisting of 50 items for e-commerce channel sales starting from 2016 is used. A well-known transformer-based Deep Learning algorithm, Temporal Fusion Transformer has been compared against numerous state-of-the-art architectures such as DeepAR, N-Beats, and NHITS and a classical time series analysis technique, ARIMA. We observe that TFT significantly outperforms benchmark algorithms and can work well with a relatively smaller dataset.

Construction industry is one of the most labor and productivity intense industries, requires a continuous improvement at every time. The basic ones are through technological destruction. In the era of Industry 4.0 and Society 5.0, focus is on machine learning, artificial intelligence and deep learning algorithms by limitless attempts. Nevertheless, in the sense of urgency, industry practitioners do not consider these approaches as much. With that ground, this paper analyzes if the terms artificial intelligence, machine learning, deep learning, internet of things (considered as digitalization), are known by construction industry practitioners. Text mining methodology through Naïve Bayes Algorithm in Turkish construction practitioners is used with lists of non-academical media in Turkish construction industry journals (n = 59). Text mining approach is based on “wordcloud” and “sentiment analysis” according to the Bayesian Theorem without classifying and pointing the emotions using R programming and corresponding packages. Model supports the hypothesis with 79

It is evident that companies invest considerable resources in creating and implementing marketing initiatives mainly to attract customers and thereby increase their market presence. Determining if these marketing efforts are effective in convincing customers to take desired actions is challenging since companies need to ensure that their marketing campaigns are feasible in order to target the right customers who are most likely to respond to the marketing treatment. Prescriptive analytics provide better means to measure the impact of a marketing campaign on customer behavior when compared to conventional predictive analytics. Uplift modeling presents an opportunity to maximize the incremental impact of marketing treatment by determining the most responsive customer segment who will take positive action only because of receiving the treatment. This paper aims to suggest an uplift modeling approach in a marketing campaign in B2B context for better evaluation of performance which has not gained enough attention as B2C in literature. By applying three uplift modeling techniques to a real-world B2B cross-sell campaign, it is demonstrated that the campaign effectiveness can be increased significantly by determining the customers who are truly responsive to the related campaign.

This paper presents a holistic NLP-based approach to the fragrance industry. Using a methodology based on online reviews, this study aimed to gain insights into customer sentiment and product preferences towards fragrance brands. In this study, analyses were conducted by creating a unique dataset by randomly selecting 13,828 reviews of 36 perfumes from a website specializing in the global perfume industry. By sentiment analysis, it is investigated how well user comments' sentiment compound scores and overall perfume evaluations correlate. Using Latent Dirichlet Allocation method, commonly used terms and topics are extracted from comments to provide insights into user sentiments. Moreover, Term Frequency-Inverse Document Frequency analysis helped to draw out keywords and word patterns from evaluations that are distinctive to a given brand. The most commonly used words in these comments were identified by examining user reviews. By bridging perfumes and language within user-generated content, we believe this study would contribute valuable insights to management practices in the fragrance industry.

Marketing departments of companies often struggle with cost-related issues arising from insufficient and inaccurate definition of their target customers. They desire to focus on customers who are highly profitable and loyal. The aim of this study is to enable companies with e-commerce activities to precisely define their target customer segments and understand their characteristics better. For this purpose, the e-commerce transaction dataset provided by a household panel company operating in Türkiye is used. In this project, where a real-life case is analysed, unsupervised machine learning algorithm, K-Means, is used. By doing cluster analysis, the eventual aim is to reach the right number of clusters having similar customer behaviours. The result demonstrates successful modelling, achieving distinct segments consisting of homogenous personas. Consequently, enterprises with e-commerce activities will be able to identify different customer types more effectively, which paves the way for the development of customised marketing strategies, increasing loyalty and profitability.

This study introduces an innovative framework that utilizes learning algorithms to rank customer reviews on e-commerce platforms. Addressing the ambiguity and subjectivity in customer feedback, our approach highlights the use of an extensive dataset and feature engineering. A pivotal part of our methodology is the creation of an original target variable named 'adjusted action rate' (AAR), combined with advanced training techniques to alleviate 'position bias'. This strategy allows us to effectively capture the nuances of user behavior and review dynamics. At the core of our framework are Learning to Rank (LTR) methods, specifically designed to tackle the unique challenges of review ranking. Our primary evaluation criterion is the Normalized Discounted Cumulative Gain (nDCG) metric, which assesses the efficiency of our LTR algorithm in predicting purchase likelihood based on user reviews. Validation through online A/B testing shows that our framework significantly improves user interaction, decision-making efficiency, and the overall shopping experience on e-commerce sites. The results confirm the success of our strategy in overcoming the complexities of review ranking, evidenced by notable enhancements in engagement metrics.

The Turkish Electricity Market has experienced various procedural transformations over time. These changes have led to the establishment of a system in the electricity market that allows stakeholders to secure hourly energy through next-day sales and purchases. This system is known as the pre-day market and the price set within this framework is referred to as the market clearing price. This study was designed to predict the electricity price for the next 24 time units within the next 24 h in Turkey. Predictions of the market clearing price were conducted using numerous machine learning models. Time series data of clearing prices in Turkey were used in the analysis. Exogenous variables such as production amount and holiday dummies were also incorporated. The data period was from January 2021 to December 2023. The study utilized two lagged market clearing price features with 19 independent lagged and unlagged additional variables. Various machine learning models were tested for their efficacy in forecasting the market clearing price, to identify the most effective one. To benefit from the various advantages of different models, the three models with the best performance, lightGBM, OMP, and STLF were blended to obtain a new model.

Due to its high cost, loss of productivity, and most importantly, loss of time in training a new employee, employee retention has become a strategy that has made even more attractive for many researchers and professionals in the field. The purpose of this study is to present a case study that addresses the problem of employee churn and develop a model which predicts employee retention best. In the present study, the most well-known machine learning techniques such as Logistic Regression, K-Nearest Neighbor (KNN), Naive Bayes, Decision Tree, Support Vector Machine (SVM), XGBoost, Artificial Neural Network (ANN) and Random Forest were used. Finally, the performance of the proposed approaches was evaluated. The numerical results showed that the proposed Naïve Bayes clearly outperformed all other classifiers according to all evaluation criteria except Accuracy. However, Random Forest gave the best results according to the accuracy criterion.

It is a matter of controversy whether the transfers made in the football industry are efficient or not. The aim of the study is to explore the efficiency of transfers made in the football industry using machine learning techniques. In this context, a methodology to model the success of transfers based on Turkish Super League data is suggested. In the modelling processes, the data of the transfers taken from the Tranfermarkt website were used. The target variable is created as binary and the classification problem is the consideration. Accordingly, the data of 16 teams and 2261 players in total were analysed using advanced machine learning methods. Results reveal that transfers of young and homegrown players are relatively more efficient compare to those of the others.

The energy sources of the universe are limited and draining with increasing speed. One of the biggest consumption of energy is in the manufacturing industry with 35% of the overall consumption. Collecting and analyzing data is becoming a crucial topic with the Internet of Things (IOT). Using the science of data is becoming a hotter topic in manufacturing industries as well. The study aimed to apply theoretical analytical methods in real-life cases and demonstrate potential returns on IOT investments for fully automated systems. This study not only analyzed and predicted the energy consumption but also provided a prescriptive receipt to minimize energy spending while keeping high-quality production standards. For the use case, the data gathered from real-production of a multinational food manufacturer and machine learning algorithms were applied. The study provides an energy efficiency of 5–6% for a single machine which consumes 35–45% of the total energy spending of the entire factory.

Sovereign credit ratings are major indicators of a country’s financial structure as they provide an assessment of the creditworthiness of a country and its capability to meet its financial obligations. On the other side, how they are established by credit rating agencies (CRAs) is considered as not transparent and objective enough. This study aims to suggest a prediction framework for sovereign credit ratings based on machine learning (ML) algorithms using the following predictors: Credit Default Swap, Government Bond Yield, GDP/Capita, Consumer Price Index, Currency Volatility, and Political Risk.

The rapid expansion of Airbnb and the increase in the number of listings on the platform have raised concerns about the impact on housing markets and urban dynamics, particularly in regards to housing affordability. This study aims to examine the relationship between Airbnb listings in Istanbul, Turkey, a major tourist destination with a high demand for accommodations and a significant housing affordability problem. Using unsupervised learning techniques, the study clusters and analyzes Airbnb listings in Istanbul to identify patterns and trends in the data. In the study, k-means and fuzzy c-means algorithms are applied to group the data into various numbers of clusters and the results were evaluated using the Silhouette Score and the Dunn Index methods. Results showed that k-means clustering algorithm could be preferred to fuzzy c-means, with the optimal number of clusters being between two and five. Authors selected k-means algorithm with five clusters for further analysis and named the resulting clusters based on their most important characteristics, providing an overview of what to expect from listings within each group, revealing valuable insights for researchers and stakeholders in the housing market.

Spot welding is a critical joining process which presents specific challenges in early defect detection, has high rework costs, and consumes excessive amounts of materials, hindering effective, sustainable production. Especially in automotive manufacturing, the welding source’s quality needs to be controlled to increase the efficiency and sustainable performance of the production lines. Using data analytics, manufacturing companies can control and predict the welding parameters causing problems related to resource quality and process performance. In this study, we aimed to define the root cause of welding defects and solve the welding input value range problem using machine learning algorithms. In an automotive production line application, we analyzed real-time IoT data and created variables regarding the best working range of welding input parameters required in the inference analysis for expulsion reduction. The results will help to provide guidelines and parameter selection approaches to model ML-based solutions for the optimization problems associated with welding.

Epics are the narratives of generally exceptional illustrations of the nation’s cultural heritage and reflect the cultural elements of the era that epic occurred. Epics are also studied by social scientists and linguists and they are about to vanish from the history. Since these are works from centuries ago, there are no images or visual records of the events. However, the detailed descriptions of characters and environments are accessible through the chapters of the epics. This study aims to suggest a framework for epic poetry summary extraction based on text mining and image generation techniques. As a result of combining the results of the used algorithms, a final output was obtained that will reveal the themes of the chapters of the epics. The outputs of the model were given as input to the text-to-image application DALL-E, and the main ideas were visualized. Results show that obtained summaries are in line with the epic literature.