Microsoft Word - Project Proposal

Amrit Science Campus Affiliated to: Tribhuvan University Institute of Science and Technology Final Year Project Proposal on COMMODITY PRICE ANALYSIS AND FORECASTING SYSTEM Submitted to: Department of Computer Science and Information Technology Amrit Campus, Lainchaur, Kathmandu In partial fulfilment of the requirement for the Bachelor Degree in Computer Science and Information Technology Submitted by: Ankit Bhusal (79010157) Aaditey Bikram Saud (79010150) Submitted Date: 2082/11/12.

ABSTRACT This project aims to analyse and forecast commodity prices using data-driven techniques. Commodity prices such as gold, crude oil, and agricultural products significantly influence economic stability, business planning, and investment decisions. The system will collect historical data, perform data pre-processing, conduct exploratory data analysis (EDA), and implement time-series forecasting models such as ARIMA and Prophet. Interactive dashboards will be developed using Power BI to present insights effectively. The expected outcome is a structured analytical framework capable of identifying trends, seasonality, and providing short- term price forecasts to support data-driven decision-making. Keywords: Commodity Forecasting, Time Series Analysis, Price Volatility, Predictive Modelling.

Contents 1. Introduction: ................................................................................................................................. 1 1.1 Background of the Study .............................................................................................................. 1 1.2 Problem Statement ........................................................................................................................ 1 1.3 Objectives ..................................................................................................................................... 1 1.4 Scope and Limitations ................................................................................................................... 1 2. Literature Review ......................................................................................................................... 2 3. Methodology ...................................................................................................................................... 3 3.1 Data Collection ............................................................................................................................. 3 3.2 Data Preprocessing ........................................................................................................................ 3 3.3 Data Analysis and Forecasting ...................................................................................................... 3 3.4 Tools and Technologies ................................................................................................................ 3 4. Feasibility Study .......................................................................................................................... 4-5 4.1 Technical Feasibility ..................................................................................................................... 4 4.2 Economic Feasibility .................................................................................................................... 4 4.3 Operational Feasibility .................................................................................................................. 5 4.4 Schedule Feasibility ...................................................................................................................... 5 4.5 Legal Feasibility ............................................................................................................................ 5 5. Algorithm Description & Technical Workflow .......................................................................... 6 6. Expected Outcome: ....................................................................................................................... 6 7. Reference ....................................................................................................................................... 7.

1 1. Introduction: Commodity markets are inherently volatile and influenced by multiple global and local factors such as geopolitical tensions, inflation, currency exchange rates, and supply-demand dynamics. Analyzing historical price data helps in understanding market behavior and anticipating future trends. With the growth of data analytics, it has become possible to transform raw financial data into valuable business intelligence. This project aims to design and implement a data- driven system to analyze and forecast commodity prices effectively. 1.1 Background of the Study In Nepal and globally, commodities such as gold and petroleum products directly impact the economic structure. Despite availability of historical data, there is limited structured analysis and forecasting accessible to stakeholders. By applying modern analytical tools and statistical techniques, more reliable and interpretable insights can be generated. 1.2 Problem Statement Current approaches to commodity monitoring rely heavily on descriptive statistics without predictive modeling. There is a lack of integrated systems that combine trend analysis, seasonality detection, correlation analysis, and forecasting in a single framework. This project aims to bridge this gap by developing a comprehensive analytical system. 1.3 Objectives • To collect and compile historical commodity price data from verified sources. • To clean and preprocess the dataset for analytical consistency. • To perform exploratory data analysis to detect patterns and trends. • To analyze seasonality, volatility, and correlations among commodities. • To implement time-series forecasting models such as ARIMA and Prophet. • To evaluate forecasting performance using statistical metrics. • To develop an interactive visualization dashboard for stakeholders. 1.4 Scope and Limitations The project focuses on selected commodities including gold, crude oil, and selected agricultural products. The forecasting will primarily cover short to medium-term predictions based on available historical datasets. The accuracy of predictions may be affected by unforeseen economic disruptions..

2 2. Literature Review Many previous studies have shown that time-series forecasting models such as ARIMA, SARIMA, and Prophet are effective for predicting financial and commodity prices. These models are designed to analyze data collected over time and identify important patterns. They break the data into key components such as trend (long-term increase or decrease), seasonality (regular repeating patterns), and residual or random variations. By understanding these components, the models can make more accurate future predictions. Researchers around the world have applied these techniques to forecast prices of commodities like gold, oil, and agricultural products. The results show that time-series models can provide reliable short-term and medium-term predictions when trained properly. However, most of these studies focus on global markets and large economies. There is comparatively less research focused on localized or country-specific commodity data. This project builds on these established forecasting techniques and applies them to a practical, real-world dataset using modern data analytics tools..

3 3. Methodology 3.1 Data Collection Data will be collected from sources such as: i. World Bank databases ii. Nepal Rastra Bank reports iii. Kaggle datasets, and other verified financial platforms. 3.2 Data Preprocessing To handle missing values, inconsistencies, and outliers, the system employs a multi-step Data Cleaning Pipeline. Missing entries are addressed through linear interpolation or mean imputation to maintain time-series continuity, while inconsistencies like duplicate records or incorrect units are standardized for uniformity. Outliers caused by data entry errors or extreme market anomalies are detected using Z-score analysis or Interquartile Range (IQR) and either capped or removed to prevent model skewing. Finally, the data is normalized (typically using Min-Max Scaling) to ensure all features reside within a specific range (0 to 1), making it mathematically compatible for training Deep Learning models like LSTM. 3.3 Data Analysis and Forecasting Exploratory Data Analysis will be conducted using Python libraries such as Pandas and Matplotlib. Forecasting models including ARIMA and Prophet will be applied. Model accuracy will be evaluated using MAE and RMSE metrics. 3.4 Tools and Technologies • Python (Pandas, NumPy, Matplotlib) • SQL • Power BI.

4 4. Feasibility Study The feasibility of the Commodity Price Analysis & Forecasting System is evaluated based on technical, economic, operational, and schedule aspects. 4.1 Technical Feasibility This study assesses whether the current technology and skills are sufficient to develop the system. Hardware Requirements: The training of deep learning models like LSTM can be computationally intensive. However, with modern laptops (8GB+ RAM, i5+ processor) or free cloud platforms like Google Colab (which provides free GPUs), the technical infrastructure is readily available. Software Stack: The project utilizes high-level, open-source libraries. Python offers robust support for machine learning via TensorFlow and Keras, while FastAPI and React provide efficient frameworks for the web interface. Data Availability: In the context of Nepal, historical data can be sourced from the Kalimati Fruits and Vegetable Market Development Board and the Nepal Oil Corporation. These datasets provide sufficient temporal depth for time-series analysis. 4.2 Economic Feasibility This project is highly cost-effective as it relies primarily on open-source technologies. Development Cost: There are no licensing fees for the programming languages (Python, JavaScript) or the ML frameworks. Operational Cost: Hosting the system on platforms like Heroku or Vercel is free for small- scale student projects. The primary "cost" is the student's time and effort. Value Proposition: If successful, the system provides significant economic value to farmers and traders by reducing financial loss through informed decision-making..

5 4.3 Operational Feasibility Operational feasibility evaluates how well the system will be used and supported by its target audience. User Interface: By developing a responsive web dashboard with interactive Plotly charts, the system ensures that even non-technical users can interpret price trends easily. Maintenance: Since the system follows a modular architecture (separate backend for ML and frontend for UI), updating the forecasting model with new algorithms in the future will be seamless. Acceptance: There is a clear demand from agricultural cooperatives and retail investors for a centralized, predictive tool that simplifies complex market data. 4.4 Schedule Feasibility The project is designed to be completed within the 9 Weeks timeframe typical of a BScCSIT final year project. 4.5 Legal Feasibility Data Privacy: The project uses publicly available market data, so there are no privacy violations. Licensing: All libraries used (MIT or Apache license) allow for educational and development use without legal restrictions. Activities Week 1-2 Week 3 Week 4 Week 5-8 Week 9 Literature Review ✔ Data Collection ✔ ✔ Data Cleaning & Preprocessing ✔ ✔ Exploratory Data Analysis ✔ ✔ Model Development & Testing ✔ ✔ Documentation & Final Presentation ✔.

6 5. Algorithm Description & Technical Workflow The system utilizes Long Short-Term Memory (LSTM), a specialized form of Recurrent Neural Network (RNN) capable of learning long-term dependencies. This is critical for commodities where prices today may be influenced by seasonal trends from several months ago. Gating Mechanisms: The LSTM cell maintains a long-term state Ct and short-term state ht. The flow of information is controlled by three main gates: 6. Expected Outcome: The project will produce well-organized analytical reports that clearly explain the patterns and behavior of commodity prices over time. These reports will include detailed statistical summaries, trend analysis, seasonal pattern identification, and correlation findings between different commodities. The forecasting results will present predicted future prices based on historical data using time-series models such as ARIMA and Prophet. These predictions will help estimate short-term and medium-term price movements. In addition to written analysis, the project will include interactive visual dashboards developed using tools like Power BI or Python visualization libraries. These dashboards will display charts such as line graphs, moving averages, seasonal decomposition plots, and comparison graphs. Users will be able to easily understand how prices fluctuate over months and years. The system will also highlight volatility and sudden changes caused by economic or global events. By combining analytical reports, forecasting models, and visual dashboards, the project will generate meaningful predictive insights. These insights can support investors, traders, policymakers, and researchers in making informed decisions. Overall, the system will transform raw commodity price data into actionable intelligence, helping stakeholders better understand market behavior and anticipate possible future trends..

7 7. Reference World Bank Commodity Price Data (https://www.worldbank.org) Nepal Rastra Bank Economic Bulletins Hyndman, R.J., & Athanasopoulos, G. (Forecasting Principles and Practice) Research articles on ARIMA and Prophet forecasting models.