DEVELOPMENT OF A FINGERPRINT-BASED AGE ESTIMATION SYSTEM TO MITIGATE UNDERAGE VOTING IN NIGERIA

Abstract

Underage voting is the participation of individuals below a certain age called voting age (18 years as obtainable in Nigeria) in public elections. It constitutes electoral fraud which is capable of undermining a nation’s democracy and create unrest during and after an election is conducted. This study designs an electoral enrolment system capable of estimating prospective electorate’s age from their fingerprints towards determination of their eligibility (identify those who does not satisfy the age requirement) using ensemble classification technique. Fingerprint data were locally sourced from certain schools in Ondo and Osun State and labelled accordingly. Fingerprints from subjects below the age of eighteen (18) were labelled Minor while those from subjects eighteen (18) years and above were labelled adult in accordance with the provisions in Nigeria’s electoral Act 2010. Data were collected from four hundred (400) subjects. Upon the removal of very dark fingerprint which could adversely affect classification results, a total of three hundred (300) fingerprints remaining constitute the design set upon which the ensemble model was developed (Trained and Tested). The fingerprints were made to undergo pre-processing operations which includes; segmentation, contrast adjustment and the removing of guassian noise using the adaptive guassian filter towards enhancing the clarity of the dataset for feature extraction. Discrete wavelet transform, support vector machine and local binary pattern where the choice feature extraction techniques, extracting the frequency domain, spatial domain and textural information of the pre-processed fingerprint. Features extracted using the above techniques were further serially concatenated to get a vector which ultimately holds information critical for training and testing of the predictive model. Three supervised machine learning algorithms (K-Nearest Neighbour, Support Vector Machine and Logistic Regression) were the predictive model of choice. Also using the three algorithms as base models, a voting ensemble classifier was developed. The implementation was done using Python programming language and its associated libraries such as Scikit Learn, Matlabplot and so on. The predictive models were trained and tested with data from the design set divided into training and testing data such that 70% of the data was dedicated to model training while the remaining 30% was used for testing. The performances of the individual models as well as that of the voting ensemble were evaluated using standard metrics such as the accuracy, Precision, Recall rate and the F1-Score. The ensemble model was observed to perform best with an accuracy of up to 0.88, slightly edging SVM and LR. The results from this study proves the efficacy of the fingerprint in age group classification task.