The gold price prediction system in Kuala Pilah, Malaysia, is an interesting area of research because the gold

price here is unpredictable. The reason for this is that the price in Kuala Pilah, Negeri Sembilan, Malaysia is a

factory price, which differs from the MS Bullion website. Predicting gold prices can be a gold mine —very

beneficial for investors, traders, and anyone who needs to plan ahead. The price of gold is constantly changing

and it cannot be easily predicted. Now that prices are rising, many customers who come to this shop are

interested in selling their gold. However, some exchange old gold for new designs to use the gold as savings

and a 'backup' to cash savings (Assan, 2023). This project aims to analyse the factors influencing gold prices in

Kuala Pilah and to develop a reliable model for predicting future price trends. Understanding the dynamics of

the gold market in this specific geographical location will enhance existing knowledge and provide practical

implications for stakeholders involved in gold trading in Kuala Pilah. This is because the gold price in Kuala

making informed decisions about buying, selling, and investing in gold. However, the price of gold fluctuates

unpredictably (Makala et. al, 2021). Current gold price prediction methods in Kuala Pilah are limited and often

To address this problem, machine learning and deep learning methods, specifically recurrent neural networks

Additionally, the problem at hand is the need for daily gold price predictions to identify optimal buying and

selling opportunities, ensuring purchases are made at lower prices and sales at higher prices (Setyowibowo et

al., 2022). This requires a solution that can accurately predict daily gold price movements. One possible

In predictive applications, LSTMs are particularly valuable for their ability to learn from historical data and

identify complex temporal patterns. For instance, in financial forecasting, an LSTM model can analyse past

stock prices to predict future values, capturing intricate dependencies that traditional models like ARIMA may

dependencies, a crucial feature for predicting outcomes in which earlier inputs significantly influence future

outputs. This makes LSTMs particularly suitable for tasks such as weather forecasting, where current

model is trained to map sequences of past observations to future values. After training, the LSTM can generate

predictions by processing the most recent data points. The model's performance is typically evaluated using

Percentage Error (MAPE). Accuracy testing validates the results produced by the developed model using the

In the gold price prediction system, a function named define_model was created to define the LSTM model.

The model was designed to accept input data with shape (128, 1), corresponding to 128 time steps with one

feature each. The model’s architecture consisted of three LSTM layers, each with 64 units. The

return_sequences=true parameter in the first two LSTM layers ensured that the entire sequence of outputs was

returned, enabling sequence predictions. After each LSTM layer, a Dropout layer with a rate of 0.2 was added

The final layer of the model was another Dense layer with a single unit, serving as the model's output and

predicting the future gold price. The model was compiled using the Nadam optimiser and mean squared error

as the loss function, both of which are suitable for a regression problem like this one. The model.summary()

function was called to print the model architecture. The function define_model then returned the constructed

and from the Bullion Rates website for Malaysia. The telegram displays the price of gold with wages, without

prepare the data for the LSTM model. This process includes handling missing values and transforming the data

into sequences for the model. The LSTM model, a type of neural network designed for time series data, is

trained on these sequences to learn historical price patterns. Finally, the trained model can be used to forecast

Figure 4 displays the use case diagram for this prototype, summarising the details of the prediction system and

error, root mean square error, mean absolute error, mean absolute percentage error, and median absolute error.

However, RMSE, MSE, and MAPE are the most popular because of their theoretical relevance in statistical

includes a feature that lets users contact the admin directly to ask questions or request an explanation of our

website. The system also helps the admin by providing visualisation of each model, along with its accuracy.

For future work on this project, the gold price data should be pulled directly from financial websites by using a

gold API. This recommendation will help improve the model’s performance and enhance system features.

Another recommendation is to enhance the predictive model by using more advanced techniques,

incorporating additional features, and gathering more historical data to improve accuracy. Regular evaluation