This study was completed as a part of an Electrical Engineering Master's thesis, with a

focus on data-driven modelling. In this study, monthly sales of electric vehicles (EVs) in Türkiye

are forecasted for the short term. Türkiye's transition to electric vehicles driven by technological

innovation, domestic electric vehicle (EV) efforts, and legislative incentives have become a

visible reality. Short-term forecasting of monthly EV sales has become essential for all parties

involved, such as automotive firms, infrastructure providers, energy planners and governments

with the development of strategies on an annual or monthly basis. Many issues such as

production, export, import, logistics, energy planning, infrastructure planning, installation and

development of charging stations or other fields benefit from forecasting. Especially for fast

developing countries such as Türkiye.

Two models are utilized for the study; multi input deep assessment model (M-DAM) and long short-term memory (LSTM) model. Multi input Deep Assessment Methodology (M-DAM), a modeling technique that best represents time series data by using fractional calculus while LSTM network is a deep learning model uses supervised learning to identify patterns in historical time-series data. Normalized monthly EV sales, hybrid vehicle sales, automobile sales, consumer price index and consumer confidence index datas were used as input for both methods covering the period from October 2016 to April 2025.

Two time-series forecasting models are first implemented and compared in this study. Preprocessing, model building, assessment, and visualisation are all included.

As a result, with a decreased MAPE (9.73%) and RMSE (0.0339), the LSTM model showed

better performance compared to M-DAM model (MAPE = 16.79% and RMSE= 0.1056%) in

fitting historical EV sales data and capturing nonlinear temporal patterns. This suggests that the

model has a great capacity to learn complicated dependencies over time. However, the M-DAM

model produced forecasts that were more accurate and more in pattern with actual values in the

short-term forecasting. Overall, the findings show a trade-off between interpretability and

adaptability: M-DAM, with its transparent structure and fewer parameters, provides useful

advantages for short-term planning and decision-making, whereas LSTM is excellent at

capturing long-range dynamics but may overfit short horizons. Hybrid techniques that combine

the advantages of both methodologies may prove advantageous for future research.