Peramalan data Time Series menggunakan Hybrid model ARIMA, Exponential Smoothing dan Long Short-Term Memory dengan Genetic Algorithm Optimization

Faturrahman, Akmal (2026) Peramalan data Time Series menggunakan Hybrid model ARIMA, Exponential Smoothing dan Long Short-Term Memory dengan Genetic Algorithm Optimization. Sarjana thesis, UIN Sunan Gunung Djati Bandung.

Preview	Text 1_cover.pdf Download (330kB) | Preview
Preview	Text 2_abstrak.pdf Download (94kB) | Preview
Preview	Text 3_skbebasplagiarism.pdf Download (264kB) | Preview
Preview	Text 4_daftarisi.pdf Download (86kB) | Preview
Preview	Text 5_bab1.pdf Download (141kB) | Preview
	Text 6_bab2.pdf Restricted to Registered users only Download (480kB) | Request a copy
	Text 7_bab3.pdf Restricted to Registered users only Download (671kB) | Request a copy
	Text 8_bab4.pdf Restricted to Registered users only Download (683kB) | Request a copy
	Text 9_bab5.pdf Restricted to Registered users only Download (80kB) | Request a copy
	Text 10_daftarpustaka.pdf Restricted to Registered users only Download (86kB) | Request a copy
	Text 11_lampiran.pdf Restricted to Repository staff only Download (285kB) | Request a copy

Abstract

The stock price movement of Bank Central Asia (BBCA) is a non-linear, non-stationary, and highly volatile time series. This research develops a parallel hybrid model that integrates the Autoregressive Integrated Moving Average (ARIMA(2,1,1)) model, Exponential Smoothing (ES) of the Error Trend Seasonal (ETS(Additive,Additive,None)) type, and Long Short-Term Memory (LSTM) with a look-back window size of 5 days, where the combination weights are optimized using a Genetic Algorithm (GA). The data used are the daily closing prices of BBCA for the period from May 28, 2018, to May 21, 2026, comprising 1,910 training data points and 90 testing data points. The GA was executed for 150 generations to find the optimal weights across three forecasting periods: short-term (7 days), medium-term (30 days), and long-term (90 days). The evaluation results indicate performance dynamics that are dependent on the prediction time period. In short-term predictions, the hybrid model yielded a higher error rate compared to the individual models, where the individual LSTM model was superior in the Mean Absolute Error (MAE) and Root Mean Square Error (RMSE) metrics, while the individual ES model excelled in the Mean Absolute Percentage Error (MAPE) metric. However, in the medium- and long-term prediction periods, the hybrid model proved to be highly robust, producing a significantly higher level of accuracy compared to all other individual models. In conclusion, this parallel hybrid architecture is proven highly effective in mitigating the weaknesses of each individual model, particularly in maintaining stability and minimizing error accumulation in medium- and long-term forecasting.

Actions (login required)

View Item