Dharmarathne, Chathuranga, Ouazzane, Karim, Shrestha, Subeksha and Fernando, Sandra (2024) Exploring machine learning models for age recognition. In: International Conference on Innovative Computing and Communication (ICICC 2024), 16-17 February 2024, Shaheed Sukhdev College of Business Studies, University of Delhi, New Delhi.
The continuous improvements of artificial intelligent models for classification and facial recognition have gotten a lot of attention in recent years, particularly in models for classifying and recognizing faces. These improvements have been instrumental in solving real-world challenges. Many studies focus on neural networks, sometimes called as "black boxes" due to their complex decision-making processes and why it forecasts a particular age. On the other hand, the accuracy of datasets, which are supposed to show a person's real age, might not be very accurate. Consequently, creating accurate and interpretable age recognition models becomes challenging. This journal about exploring a machine learning models that predicts a person's age based on facial images, using both traditional machine learning techniques and neural networks. The study utilizes a facial image dataset from the Imperial College of the United Kingdom. Image features are extracted using the Cannes edges method, and Principal Component Analysis reduces dataset dimensions. Random Forest, Support Vector Machine, Decision Tree, and Convolutional Neural Network (CNN), are applied to identify the most accurate model to predict age of the person. Performance metrics, confusion matrix, accuracy curve and loss curve were used to evaluate models. The model built using random forest algorithm was the highest accurate model of 60% and the SVM was recorded least accuracy of 49% compared to other models. On the other hand, CNN without convolution layer has performed 71% accuracy against 54% accuracy.
Restricted to Repository staff only until 1 August 2025.
Download (870kB) | Request a copy
SS-023.854d74dda6a3a3f61ed0.pdf
Restricted to Repository staff only
Download (181kB) | Request a copy
View Item |