An AI-Driven Framework for Anomaly Detection and Data Correction Using Frequent Activity Pattern Learning

Abstract

With the increasing adoption of smart environments and sensor-based intelligent systems, large volumes of behavioral activity data are continuously generated. Such datasets often contain unusual patterns caused by abnormal user behavior, system malfunctions, environmental variations, or data inconsistencies. Traditional monitoring approaches, which rely on manual inspection or rule-based techniques, are often inadequate for identifying complex anomalies present in large-scale and high-dimensional behavioral datasets.This project proposes a machine learning-based automated anomaly detection framework designed to identify abnormal patterns in behavioral sensor activity data. The proposed system utilizes multiple anomaly detection techniques, including Isolation Forest, One-Class Support Vector Machine (OC-SVM), Autoencoder, and LSTM-based sequence anomaly detection. The dataset is preprocessed to extract meaningful behavioral features such as sensor identification, activity state, time of occurrence, and day-wise activity patterns. These features enable the models to learn normal behavioral trends and assign anomaly scores to individual events. Events with significantly high anomaly scores are classified as abnormal. Experimental analysis demonstrates that the proposed framework provides an efficient, scalable, and intelligent solution for anomaly detection in smart environments and large behavioral datasets.