[Paper Review] Fixing Data Anomalies with Prediction Based Algorithm in Wireless Sensor Networks

Abhishek Kr. Singh, Bollibisai Giridhar+1 more|arXiv (Cornell University)|Nov 14, 2011

Energy Efficient Wireless Sensor Networks4 references5 citations

TL;DR

This paper proposes an ARIMA-based prediction algorithm to detect and correct data anomalies in wireless sensor networks (WSNs) using only individual node history. By fitting an ARIMA(p,d,q) model to each sensor's time series data at the sink, the method forecasts values to replace anomalous readings within a 95% confidence interval, achieving a maximum error of 3.62% over five forecast steps, thus improving data accuracy without relying on spatial or temporal correlations between nodes.

ABSTRACT

Data inconsistencies are present in the data collected over a large wireless sensor network (WSN), usually deployed for any kind of monitoring applications. Before passing this data to some WSN applications for decision making, it is necessary to ensure that the data received are clean and accurate. In this paper, we have used a statistical tool to examine the past data to fit in a highly sophisticated prediction model i.e., ARIMA for a given sensor node and with this, the model corrects the data using forecast value if any data anomaly exists there. Another scheme is also proposed for detecting data anomaly at sink among the aggregated data in the data are received from a particular sensor node. The effectiveness of our methods are validated by data collected over a real WSN application consisting of Crossbow IRIS Motes \cite{Crossbow:2009}.

Motivation & Objective

• To address data anomalies in wireless sensor networks caused by faulty sensors, transmission loss, or sampling errors.
• To develop a lightweight, node-specific method for detecting and correcting anomalous data without relying on spatial or temporal correlations among nodes.
• To validate the effectiveness of ARIMA modeling for real-time anomaly correction in WSNs using empirical data from Crossbow IRIS Motes.
• To provide a fault-tolerant mechanism that enables data reconstruction during transient sensor failures using historical data.
• To establish a generalized framework for anomaly detection and correction that can be applied at the sink with minimal computational overhead.

Proposed method

• A statistical ARIMA(p,d,q) model is fitted to the historical time series data of each individual sensor node at the sink to capture temporal autocorrelation and trends.
• The ARIMA model generates forecasted values for future data points, which are used to detect anomalies when actual values fall outside a 95% prediction interval.
• Anomalies are corrected by replacing outlying values with the corresponding forecasted values from the ARIMA model.
• The method uses a 95% confidence interval to define upper and lower bounds for anomaly detection, with tighter bounds possible by adjusting the confidence level.
• The model is trained and validated on real WSN data collected from Crossbow IRIS Motes over an extended period, ensuring practical relevance.
• The framework is fully decentralized per node, requiring only local data—no inter-node correlation or contextual information is used.

Experimental results

Research questions

• RQ1 Can ARIMA modeling effectively detect and correct data anomalies in individual sensor nodes using only their historical data?
• RQ2 What is the accuracy of ARIMA-based forecasting in correcting anomalous sensor readings within a real WSN deployment?
• RQ3 How does the prediction error evolve with increasing forecast horizons, and what is the optimal forecast window for reliable correction?
• RQ4 To what extent can this method reduce false positives in anomaly detection while maintaining data integrity?
• RQ5 Can this approach serve as a fault-tolerant mechanism during transient sensor failures by reconstructing missing or corrupted data?

Key findings

• The ARIMA model achieved a maximum percentage error of 3.62% when correcting anomalies over the first five forecast steps, indicating high accuracy in data recovery.
• Prediction error increased with forecast horizon, becoming significant beyond 10 steps, suggesting that forecasts within 5–10 steps are most reliable for anomaly correction.
• The 95% confidence interval used in Algorithm 2 effectively identified anomalous data points, with a Type I error rate of 5%, corresponding to the chosen significance level.
• The method successfully corrected anomalous data without requiring spatial or temporal correlation between nodes, proving its independence and scalability.
• The framework demonstrated robustness on real-world data from Crossbow IRIS Motes, validating its practical applicability in real WSN deployments.
• The study confirms that ARIMA is a lightweight yet powerful statistical tool for anomaly detection and correction in WSN data streams, especially when computational resources are constrained at the sink.