Prediksi Jumlah Pencari Kerja Menggunakan Long Short-Term Memory Berbasis Time Series pada Dinas Ketenagakerjaan Kota Pematangsiantar

Abstract

The number of job seekers is an important indicator for planning regional employment services. This study aims to develop a forecasting model for the number of job seekers at the Manpower Office of Pematangsiantar City using Long Short-Term Memory (LSTM) based on time series data. The dataset consists of monthly data from 2020 to 2025, which were aggregated into annual records with the variables of total job seekers, male job seekers, female job seekers, senior high school graduates, vocational high school graduates, and bachelor degree graduates. The research stages include data collection, preprocessing, Min-Max Scaling normalization, sequence formation using a three-year time step, LSTM model training, and evaluation using Mean Absolute Error (MAE). The best model used one LSTM layer with 64 units, Dense layers with 32 and 16 neurons, a linear output layer, and the Adam optimizer with a learning rate of 0.0005. The testing results for the 2023-2025 period show that the predicted values are very close to the actual data, namely 305.90 for 2023, 377.17 for 2024, and 410.88 for 2025, with an MAE of 0.1293 and an average prediction closeness of 99.96%. The forecast for 2026 estimates 334 job seekers. These results indicate that LSTM can be used as a decision-support approach for employment service planning; however, the results should be interpreted carefully because the number of annual observations remains limited.