HIDAYATULLAH, Muhammad Syarif and Syafei, Wahyul Amien and Tarno, Tarno (2026) EVALUASI KUALITAS COUNTERFACTUAL EXPLANATIONS DALAM INTERPRETASI SISTEM FORECASTING BEBAN LISTRIK BERBASIS RANDOM FOREST REGRESSION. Masters thesis, UNIVERSITAS DIPONEGORO.

	Text cover aja.pdf Download (136kB)
	Text Cover.pdf Restricted to Repository staff only Download (924kB)
	Text Bab 1.pdf Download (263kB)
	Text Bab 2.pdf Download (1MB)
	Text Bab 3.pdf Restricted to Repository staff only Download (1MB)
	Text Bab 4.pdf Restricted to Repository staff only Download (1MB)
	Text Bab 5.pdf Restricted to Repository staff only Download (234kB)
	Text Daftar Pustaka.pdf Download (269kB)

Abstract

Forecasting beban listrik merupakan komponen penting dalam pengelolaan sistem energi karena memengaruhi perencanaan operasional, efisiensi produksi energi, dan stabilitas jaringan. Berbagai model machine learning telah digunakan untuk meningkatkan akurasi prediksi, salah satunya Random Forest Regression. Namun, interpretasi model ini umumnya masih terbatas pada Feature Importance yang bersifat global dan statis, serta belum mampu menjelaskan alasan spesifik di balik prediksi pada tingkat individual. Penelitian ini bertujuan melakukan evaluasi terhadap kualitas Counterfactual Explanations dalam interpretasi sistem forecasting beban listrik berbasis Random Forest Regression. Dataset yang digunakan berupa data historis beban listrik per jam dari negara Panama periode tahun 2015–2019 yang diperkaya dengan variabel meteorologi dan kalender. Model dilatih menggunakan algoritma Random Forest Regression. Framework Diverse Counterfactual Explanations (DiCE) digunakan untuk membangun skenario Counterfactual Explanations dengan target penurunan prediksi sebesar 20% pada setiap instance yang memiliki beban puncak tertinggi. Kualitas penjelasan dievaluasi menggunakan metrik validity, proximity, dan compactness. Hasil penelitian menunjukkan bahwa Counterfactual Explanations mampu menghasilkan skenario perubahan minimal pada fitur input yang efektif menggeser hasil prediksi model. Temuan ini menunjukkan bahwa pendekatan tersebut dapat meningkatkan interpretabilitas sistem forecasting serta mendukung eksplorasi skenario “what-if” yang lebih informatif bagi pengambilan keputusan.
Kata Kunci : Forecasting Beban Listrik, Random Forest Regression, Explainable AI, Counterfactual Explanations, DiCE, Interpretabilitas Model

Electricity load forecasting is an essential component in energy system management as it influences operational planning, energy production efficiency, and power grid stability. Various machine learning models have been applied to improve prediction accuracy, one of which is Random Forest Regression. However, the interpretability of this model is generally limited to Feature Importance, which provides a global and static perspective and is unable to explain the specific reasons behind predictions at the individual instance level. This study aims to evaluate the quality of Counterfactual Explanations in interpreting an electricity load forecasting system based on Random Forest Regression. The dataset used consists of hourly historical electricity load data from Panama covering the period 2015–2019, enriched with meteorological and calendar-related variables. The forecasting model was developed using the Random Forest Regression algorithm. The Diverse Counterfactual Explanations (DiCE) framework was employed to generate counterfactual scenarios with a target reduction of 20% in the predicted value for instances with the highest peak load. The quality of the explanations was evaluated using validity, proximity, and compactness metrics. The results indicate that Counterfactual Explanations can generate minimal changes in input features that effectively shift the model’s prediction outcomes. These findings demonstrate that the proposed approach can enhance the interpretability of forecasting systems and support the exploration of more informative “what-if” scenarios for decision-making.
Keywords : Electricity Load Forecasting, Random Forest Regression, Explainable AI, Counterfactual Explanations, DiCE, Model Interpretability

Actions (login required)

View Item