Leveraging Statistical Analysis and Interactive Dashboards for Diesel Consumption Monitoring in United Nations Peacekeeping Operations
Summary
No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.
Summarise with VIRAT Free to try with your Science Catalogs account.Abstract
Diesel fuel is a critical and costly resource in operations that rely on vehicle fleets and stationary generators, particularly in mission-based environments where efficiency and accountability are essential. Although IoT-based fuel monitoring systems enable continuous, high-frequency data collection, the resulting large and complex datasets are difficult to interpret using conventional reporting tools, limiting their practical value for operational decision-making. This thesis develops a structured analytical framework, guided by the Cross-Industry Standard Process for Data Mining (CRISP-DM), for IoT-based diesel fuel consumption monitoring using data collected from multiple storage tanks operated by the United Nations Interim Administration Mission in Kosovo (UNMIK). The proposed approach emphasizes robust preprocessing to transform raw sensor measurements into reliable consumption time series, followed by exploratory analysis using Seasonal–Trend decomposition by Loess (STL) to identify long-term trends, seasonal behavior, and irregular variations. Spatial autocorrelation analysis using Global Moran’s I is applied to assess whether geographic proximity contributes additional explanatory information. Short-term fuel consumption forecasting is conducted over a 30-day horizon using a baseline statistical model and multiple machine learning and deep learning approaches, including LightGBM, XGBoost, Random Forest, and Long Short-Term Memory (LSTM) networks. The findings demonstrate that systematic preprocessing substantially enhances the reliability and interpretability of subsequent analyses. Spatial dependence across tanks was found to be weak, supporting independent per-tank modeling strategies. Forecasting evaluation reveals a clear performance hierarchy: the statistical Exponential Smoothing (ETS) baseline achieved an R2 of 0.650, whereas tree-based ensemble models substantially improved predictive accuracy, achieving R2 values above 0.91. The LSTM model delivered the strongest overall performance, reaching an R2 of 0.929 and reducing RMSE by approximately 55% compared to the statistical baseline. The analytical outputs are deployed through an interactive Power BI dashboard that consolidates preprocessing outcomes, temporal consumption patterns, and forecasting results into an accessible decision-support tool. Overall, this thesis provides a practical and reusable framework for transforming high-volume IoT fuel data into actionable insights for operational monitoring and planning.
Fields of Study & Subjects
Time Series Analysis and Forecasting
Forecasting Techniques and Applications
Data Mining and Machine Learning Applications
Publication details
- Journal / source
- Repositori UJI (Universitat Jaume I)
- Publication date
- 3 March 2026
- Language
- EN
- Document type
- Thesis