Academic Project · TensorFlow · Keras · Python · Google Colab
<aside> 💡
This project implements deep learning models to solve two real-world problems — fruit image classification for agriculture and stock price forecasting for finance — as part of the Deep Learning module at UTMSPACE.
</aside>
Built two deep learning models covering the full pipeline from data preprocessing and exploratory data analysis (EDA) to model training, evaluation and visualisation. Both models were developed using TensorFlow and Keras in Google Colab, under the context of ABC Corporation which aims to apply deep learning to optimise operations for its agricultural and financial sector clients.
Traditional fruit sorting in packaging and processing lines relies heavily on manual human inspection. This is time consuming, prone to human error, and difficult to scale. Misclassification of fruits leads to quality control issues and increased operational costs.
<aside> 💡
This project addresses this by building a CNN model that automatically classifies fruit images — specifically apples and bananas — simulating an automated sorting mechanism that reduces human error and improves efficiency in agricultural processing lines.
</aside>
Financial markets are highly volatile and unpredictable. Making accurate investment decisions requires analysing large volumes of historical market data, which is difficult and time consuming to do manually. Inaccurate predictions can lead to significant financial losses.
<aside> 💡
This project addresses this by building RNN and LSTM models that learn from historical stock price data to forecast future prices — enabling more accurate and data driven financial decision making.
</aside>
A Convolutional Neural Network (CNN) was built to classify images of fruits — specifically apples and bananas. The model was trained on a dataset of images organised into train and validation directories, with images standardised to 200x200 pixels and pixel values normalised to a range of 0 to 1.
Images were extracted from ZIP archives and organised into structured directories — train and validation — each containing apple and banana subdirectories. A 4x4 grid of sample images was visualised to verify data quality before training