SOCIAL MEDIA ADDICTION PREDICTION USING MACHINE LEARNING TECHNIQUES

Nowadays, social media is seen as a typical way to kill time. Concerns over social media's effects on productivity,mental health, and general well-being has grown significantly as a result of its explosive expansion. Overusefrequently results in social media addiction, a behavioral disease marked by excessive use and a loss of self-control.Using demographic and behavioral data, this study presents the design, implementation, and evaluation of amachine learning prediction model to forecast the degree of social media addiction. The research employed adataset obtained from a public accessible online digital repository called Kaggle containing important features likeage, frequency of use, happiness level, productivity loss, and self-control. The dataset collected was subjected to acomprehensive preprocessing, including data cleaning, normalization, encoding, feature scaling, outlier analysis,feature engineering and further pre-processed based on the Machine learning design pipeline. Prior to being trained,four (4) machine learning algorithms, including Random Forest (RF), Logistic Regression, Support (LR) VectorClassifier (SVC), and K-Nearest Neighbors (KNN) classifiers, were implemented and comparatively evaluated.Performance metrics such as accuracy, precision, recall, F1-score, ROC_AUC were used for assessment. Resultsindicated that the K-Nearest Neighbor achieved the best realistic performance than Random Forest, LogisticRegression, and SVC models, which all gave performance figures that were almost flawless in terms of the metricswith an accuracy of 87% and ROC_AUC of 0.97. The biggest predictors of addiction, according to featureimportance analysis, were self-control, productivity loss, frequency of use, and satisfaction level. The result alsoshowed that machine learning has the potential to be a useful predictive tool for the early identification of socialmedia addiction. However, the results' generalizability was limited by bias in the sample and theabsence of representation from African consumers. The study suggests developing real-time monitoring tools,integrating psychological and environmental aspects, and using more diversified datasets.