Published Papers

The aim of this paper is to identify the potential of using Machine Learning algorithms to predict exoplanet habitability through publicly available datasets. An initial dataset of more than 5000 observations was analyzed using Logistic Regression and Random Forest classifiers. Out of the two Machine Learning models, Logistic Regression performed poorly due to high class imbalance whereas the Random Forest classifier achieved high accuracy with strong specificity and sensitivity. The dataset was later expanded to include more than 10,000 confirmed exoplanets, with the habitability parameter redefined via the Earth Similarity Index. This change modified the percentage of habitable exoplanets to decrease from around 4% to less than 0.7%. While the Random Forest model achieved perfect accuracy, potential overfitting was noted. 

This study constructs a nonlinear system of ordinary differential equations to examine the interplay among inflation, the exchange rate (USD/TRY), and the policy interest rate in Turkey. The framework incorporates quadratic terms to capture accelerating inflation and depreciation dynamics. The coefficients are calibrated using monthly Turkish macroeconomic data via least-squares estimation with heteroskedasticity-robust errors. The system is linearised around a single equilibrium point at approximately (π*, e*, i*) = (1.27, 2.15, 0.47) to analyse its local behaviour. The Jacobian of the linearised system is used to compute its eigenvalues, which reveal instability due to a positive real eigenvalue. Simulations show saddle-type behaviour in which inflation diverges away from equilibrium. Counterfactual analysis examines whether stronger policy responses could stabilise the system, but the dominant eigenvalue remains positive across the entire parameter range, indicating that no feasible action by the Central Bank of the Republic of Türkiye (CBRT) can ensure local stability. These findings suggest that Turkey’s persistent price pressures stem from nonlinear dynamics that cannot be fully addressed through standard monetary instruments, demonstrating the value of nonlinear dynamic modelling for interpreting unstable macroeconomic environments. 

'Fast Fashion' is a term used to define the sub-section of the clothing industry that produces cheap, low-quality, and trendy clothing (often, they are replicas of high-fashion items) in a very short amount of time, catering to the wants of the customer at the point of highest demand at a very low price. To quantify the increase in clothing purchases, in 1980, an average American bought 12 items of clothing annually, but today, the number is up to 68. This relatively new concept is taking the industry by storm financially with extraordinary growth rates. It is also impacting the environment negatively with a high and irresponsible consumption of natural resources in the economy. The labor conditions in the fashion world are also worsening with the need for faster production cycles stressing underpaid labor in developing countries. This paper discusses fast fashion, its origins, and the serious economic effect it is having today.

Neurodegenerative disorders such as Alzheimer’s and Parkinson’s are challenging to detect early due to their gradual onset. This study investigates the use of machine learning algorithms to identify these conditions based on phonetic features in speech. By analyzing vocal attributes, such as fluency, articulation, and acoustic variation; this research aims to establish non-invasive diagnostic models. Principal Component Analysis (PCA) was used for feature selection, while Random Forest and Support Vector Machine (SVM) classifiers were deployed for detection accuracy. Results show promising accuracy levels, particularly in the Alzheimer’s model, highlighting the potential of AI in enhancing early clinical screening for cognitive decline. 

This case series examined the potential cognitive and neuro-physiological effects of daily auditory stimulation at a 40 Hz gamma frequency with dementia patients. In total, twenty older adults, ages 65–84 years and clinically diagnosed with mild to moderate dementia, completed a structural auditory stimulation procedure lasting for 15 minutes per day for 30 consecutive days. Assessments of cognition were completed using the standardized Mini-Cog test, and resulting neural responses  were quantitatively examined  with electroencephalography (EEG), focusing on gamma-band oscillatory activity. The results demonstrated 40% of participants showed statistically relevant improvements over pre stimulation Mini-Cog scores, demonstrating improvements in memory, attention, and executive functioning.

The relentless pursuit of alpha has driven the creation of several quantitative trading strategies, including pairs trading. The recent emergence of reinforcement learning frameworks, such as Q-learning, has led to the development of advanced statistical arbitrage models. This study presents an empirical investigation into the application of ensemble Q- learning, a reinforcement learning method, to pairs trading in the Indian equity market. Utilising hourly data from the NIFTY 50 index constituents, which covers over 128,400 data points, a novel composite scoring framework was devised to select optimal asset pairs, balancing both long-term cointegration and mean-reversion criteria. A custom trading environment was constructed using OpenAI’s gym library to realistically simulate capital constraints, transaction costs, and position management. The agent maintained multiple independently trained Q-tables, aggregating their estimates to reduce variance and improve stability in trading decisions. Hyperparameters for the agent were optimised via a Bayesian search, and strategy performance was evaluated on out-of-sample data across multiple independent runs. Results demonstrate that the ensemble Q-learning agent delivered an average six-month portfolio return of 105.2% with a Sharpe ratio of 2.08, substantially outperforming both traditional pairs trading benchmarks and major global indices. These findings provide rigorous evidence that reinforcement learning, when coupled with well-designed pairs trading strategies, can drive superior alpha generation in financial markets.