Estimation of Unemployment Rates through Machine Learning: An Application for Türkiye

Gülferah Ertürkmen; Ali Öter

Makale detayları

Estimation of Unemployment Rates through Machine Learning: An Application for Türkiye

Yazarlar:

Gülferah Ertürkmen | Ali Öter

Yayıncı:

Mustafa Süleyman ÖZCAN

Ana Başlıklar:

Makro İktisat (Diğer)

Dergi:

İnsan ve Toplum Bilimleri Araştırmaları Dergisi Vol.14 Issue 2, pp.869-886

Belge Tipi:

Makale

Yayın Tarihi:

6/2025

Belge Dili:

Türkçe

Document ID (Org No.):

2001349622

pISSN:

2147-1185

Anahtar kelimeler:

Özet:

 This research compares and analyzes traditional statistical methods and machine learning techniques for forecasting unemployment rates in Türkiye. Unemployment rates are affected by macroeconomic variables such as economic growth, inflation, population growth, migration movements and education expenditures. Therefore, unemployment rates are estimated using machine learning algorithms such as Random Forests RF, Gradient Boosting GB and Multilayer Perceptron MLP based on TurkStat data and the performances of the models are compared. In the research, among the machine learning models, the MLP model showed the best forecasting performance MAE: 1,945; RMSE: 2,235 with the lowest error rates. Although the RF and GB models achieved a certain level of accuracy, their error rates were higher compared to the MLP model. The findings suggest that machine learning techniques are more successful in unemployment forecasting than traditional statistical models. In particular, the MLP model provides more accurate forecasts than other models thanks to its capacity to learn nonlinear relationships. Moreover, correlation analyses reveal that unemployment rates are significantly correlated with inflation, economic growth and migration flows. While the negative effect of economic growth on unemployment is clearly observed, migration movements are found to increase unemployment rates. In particular, the negative correlation between inflation and education expenditures suggests that education investments decrease during periods of economic instability. In this research, unemployment rates for the years 2025, 2026 and 2027 are estimated using machine learning techniques. As a result of the analysis, the unemployment rate is expected to vary between 9.2 and 11.5 in 2025, while this rate is expected to be between 8.8 and 11.0 in 2026. According to the 2027 forecasts, the unemployment rate is expected to decline to between 8.5 and 10.7. The results of the analysis show that the MLP model provides the closest forecasts to the TurkStat data. The RF and GB models, on the other hand, have a wider margin of error, predicting unemployment rates in the range of 9-14. These forecasts are an important guide for economic policy makers and labor market analysts and provide predictions about the future course of unemployment rates.

 This research compares and analyzes traditional statistical methods and machine learning techniques for forecasting unemployment rates in Türkiye. Unemployment rates are affected by macroeconomic variables such as economic growth, inflation, population growth, migration movements and education expenditures. Therefore, unemployment rates are estimated using machine learning algorithms such as Random Forests RF, Gradient Boosting GB and Multilayer Perceptron MLP based on TurkStat data and the performances of the models are compared. In the research, among the machine learning models, the MLP model showed the best forecasting performance MAE: 1,945; RMSE: 2,235 with the lowest error rates. Although the RF and GB models achieved a certain level of accuracy, their error rates were higher compared to the MLP model. The findings suggest that machine learning techniques are more successful in unemployment forecasting than traditional statistical models. In particular, the MLP model provides more accurate forecasts than other models thanks to its capacity to learn nonlinear relationships. Moreover, correlation analyses reveal that unemployment rates are significantly correlated with inflation, economic growth and migration flows. While the negative effect of economic growth on unemployment is clearly observed, migration movements are found to increase unemployment rates. In particular, the negative correlation between inflation and education expenditures suggests that education investments decrease during periods of economic instability. In this research, unemployment rates for the years 2025, 2026 and 2027 are estimated using machine learning techniques. As a result of the analysis, the unemployment rate is expected to vary between 9.2 and 11.5 in 2025, while this rate is expected to be between 8.8 and 11.0 in 2026. According to the 2027 forecasts, the unemployment rate is expected to decline to between 8.5 and 10.7. The results of the analysis show that the MLP model provides the closest forecasts to the TurkStat data. The RF and GB models, on the other hand, have a wider margin of error, predicting unemployment rates in the range of 9-14. These forecasts are an important guide for economic policy makers and labor market analysts and provide predictions about the future course of unemployment rates.

 This research compares and analyzes traditional statistical methods and machine learning techniques for forecasting unemployment rates in Türkiye. Unemployment rates are affected by macroeconomic variables such as economic growth, inflation, population growth, migration movements and education expenditures. Therefore, unemployment rates are estimated using machine learning algorithms such as Random Forests RF, Gradient Boosting GB and Multilayer Perceptron MLP based on TurkStat data and the performances of the models are compared. In the research, among the machine learning models, the MLP model showed the best forecasting performance MAE: 1,945; RMSE: 2,235 with the lowest error rates. Although the RF and GB models achieved a certain level of accuracy, their error rates were higher compared to the MLP model. The findings suggest that machine learning techniques are more successful in unemployment forecasting than traditional statistical models. In particular, the MLP model provides more accurate forecasts than other models thanks to its capacity to learn nonlinear relationships. Moreover, correlation analyses reveal that unemployment rates are significantly correlated with inflation, economic growth and migration flows. While the negative effect of economic growth on unemployment is clearly observed, migration movements are found to increase unemployment rates. In particular, the negative correlation between inflation and education expenditures suggests that education investments decrease during periods of economic instability. In this research, unemployment rates for the years 2025, 2026 and 2027 are estimated using machine learning techniques. As a result of the analysis, the unemployment rate is expected to vary between 9.2 and 11.5 in 2025, while this rate is expected to be between 8.8 and 11.0 in 2026. According to the 2027 forecasts, the unemployment rate is expected to decline to between 8.5 and 10.7. The results of the analysis show that the MLP model provides the closest forecasts to the TurkStat data. The RF and GB models, on the other hand, have a wider margin of error, predicting unemployment rates in the range of 9-14. These forecasts are an important guide for economic policy makers and labor market analysts and provide predictions about the future course of unemployment rates.

 Bu çalışma, Türkiyede işsizlik oranlarının tahmini için geleneksel istatistiksel yöntemler ile makine öğrenmesi tekniklerini karşılaştırarak analiz etmektedir. İşsizlik oranları; ekonomik büyüme, enflasyon, nüfus artışı, göç hareketleri ve eğitim harcamaları gibi makroekonomik değişkenlerden etkilenmektedir. Bu nedenle, TÜİK verilerine dayalı olarak Rastgele Ormanlar RF, Gradyan Artırma GB ve Çok Katmanlı Algılayıcı MLP gibi makine öğrenmesi algoritmaları kullanılarak işsizlik oranları tahmin edilmiş ve modellerin performansları karşılaştırılmıştır. Çalışmada, makine öğrenmesi modelleri arasından MLP modeli en düşük hata oranları ile en iyi tahmin performansını MAE: 1,945, RMSE: 2,235 göstermiştir. RF ve GB modelleri belirli bir doğruluk seviyesine ulaşsa da, hata oranları MLP modeline kıyasla daha yüksek çıkmıştır. Bulgular, makine öğrenmesi tekniklerinin işsizlik tahmininde geleneksel istatistiksel modellere göre daha başarılı olduğunu göstermektedir. Özellikle MLP modeli, doğrusal olmayan ilişkileri öğrenebilme kapasitesi sayesinde diğer modellere göre daha hassas tahminler sunmuştur. Ayrıca, yapılan korelasyon analizleri, işsizlik oranlarının enflasyon, ekonomik büyüme ve göç hareketleri ile anlamlı ilişkiler içinde olduğunu ortaya koymuştur. Ekonomik büyümenin işsizlik üzerindeki negatif etkisi belirgin şekilde gözlemlenirken, göç hareketlerinin işsizlik oranlarını artırabileceği tespit edilmiştir. Özellikle enflasyon ve eğitim harcamaları arasındaki negatif korelasyon, ekonomik istikrarsızlık dönemlerinde eğitim yatırımlarının azaldığını göstermektedir. Bu çalışmada, makine öğrenmesi teknikleri kullanılarak 2025, 2026 ve 2027 yıllarına ait işsizlik oranları tahmin edilmiştir. Yapılan analizler sonucunda, 2025 yılında işsizlik oranının 9,2 ile 11,5 arasında değişmesi beklenirken, 2026 yılında bu oranın 8,8 ile 11,0 seviyelerinde olacağı öngörülmüştür. 2027 yılı tahminlerine göre ise işsizlik oranının 8,5 ile 10,7 seviyelerine gerileyeceği hesaplanmıştır. Analiz sonuçları, MLP modelinin TÜİK verilerine en yakın tahminleri sunduğunu göstermektedir. RF ve GB modelleri ise işsizlik oranlarını 9-14 aralığında tahmin ederek daha geniş bir hata payına sahip olmuştur. Bu tahminler, ekonomik politika yapıcıları ve işgücü piyasası analistleri için önemli bir rehber niteliğinde olup, işsizlik oranlarının gelecekteki seyri hakkında öngörüler sunmaktadır.

Makale detayları

Estimation of Unemployment Rates through Machine Learning: An Application for Türkiye

Yazarlar:

Gülferah Ertürkmen | Ali Öter

Yayıncı:

Mustafa Süleyman ÖZCAN

Ana Başlıklar:

Makro İktisat (Diğer)

Dergi:

İnsan ve Toplum Bilimleri Araştırmaları Dergisi Vol.14 Issue 2, pp.869-886

Belge Tipi:

Makale

Yayın Tarihi:

6/2025

Belge Dili:

Türkçe

Document ID (Org No.):

2001349622

pISSN:

2147-1185

Unemployment

Machine Learning

Unemployment Rate Prediction

Random Forests

Gradient Boosting

Multilayer Perceptronİşsizlik

Makine Öğrenmesi

İşsizlik Oranı Tahmini

Rastgele Ormanlar

Gradyan Artırma

Çok Katmanlı Algılayıcı

Sayı

Makale detayları

Yazarlar: Gülferah Ertürkmen | Ali Öter

Yayıncı: Mustafa Süleyman ÖZCAN

Ana Başlıklar: Makro İktisat (Diğer)

Dergi: İnsan ve Toplum Bilimleri Araştırmaları Dergisi Vol.14 Issue 2, pp.869-886

Belge Tipi: Makale

Yayın Tarihi: 6/2025

Belge Dili: Türkçe

Document ID (Org No.): 2001349622

pISSN: 2147-1185

Sayı

Yazarlar:

Gülferah Ertürkmen | Ali Öter

Yayıncı:

Mustafa Süleyman ÖZCAN

Ana Başlıklar:

Makro İktisat (Diğer)

Dergi:

İnsan ve Toplum Bilimleri Araştırmaları Dergisi Vol.14 Issue 2, pp.869-886

Belge Tipi:

Makale

Yayın Tarihi:

6/2025

Belge Dili:

Türkçe

Document ID (Org No.):

2001349622

pISSN:

2147-1185