Volume 10, Issue 3, March – 2025 International Journal of Innovative Science and Research Technology

ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/25mar233

Anomaly Detection in Kitchen Appliance

Usage using Machine Learning Approach
Javid Gahramanov1; Naeem Naseer2; Ayishagul Gahramanova3
1
Software Architect, JVD Smart Systems, Baku, Azerbaijan
2
Lecturer, Muhammad Nawaz Sharif University of Agriculture, Multan, Pakistan
3
Mathematician Baku, Azerbaijan

Publication Date: 2025/03/20

Abstract: Anomaly detection in domestic energy consumption is crucial for improving energy efficiency, identifying faulty
appliances, and identifying unusual patterns of consumption. In this paper, the Isolation Forest (IF) algorithm is employed
to detect anomalies in domestic appliance power consumption. The data consists of high-resolution (1Hz) domestic appliance
power measurements with three chosen features: DISHWASH, HEATHOME, and AIRCOND, corresponding to dishwasher
consumption, home heating, and air conditioning, respectively. These features were selected because they are major
contributors to overall energy usage and have the ability to point towards unusual trends. Preprocessing involved
normalizing the selected features and using Isolation Forest for anomaly detection. IF separates anomalies by recursive
partitioning, effectively separating outliers from regular data. KDE analysis identified that all three features have a bimodal
distribution, which indicates different consumption patterns. Pair plot and 3D visualization outcomes verify that IF
accurately detected two separate outliers. Further, the histogram of IF anomaly scores verify these results, wherein a 5%
contamination level is utilized to distinguish normal and anomalous points. The findings show that unsupervised machine
learning models such as Isolation Forest are able to identify effectively anomalies in household energy consumption,
providing information on potential energy losses and system errors. The study aims to enhance smart energy monitoring
systems through predictive maintenance and energy optimization applications.

Keywords: Anomaly Detection, Isolation Forest, Energy Consumption, Machine Learning, Smart Home Monitoring.

How to Cite: Javid Gahramanov; Naeem Naseer; Ayishagul Gahramanova (2025). Anomaly Detection in Kitchen Appliance Usage
using Machine Learning Approach. International Journal of Innovative Science and Research Technology, 10(3), 425-430.
https://doi.org/10.38124/ijisrt/25mar233

I. INTRODUCTION A typical anomaly detection system comprises two

major steps. The first step is to collect data from the real-
The past few years, technologies in smart home sensors world environment with sensors, either vision-based or
have developed significantly and have been able to capture sensor-based, depending on the domain [5]. The second phase
and exchange information more efficiently. Applications of is the behavior analysis phase where anomalous behaviors are
artificial intelligence and machine learning (ML) have made identified and proper ML methodologies are utilized for
the outcomes more accurate and solid. This has enabled a lot distinguishing normal behavior from anomalous behavior.
of applications to perform intricate calculations with ease. Anomalous behavior commonly arises when an activity is
One of them is human activity recognition (HAR), undertaken at an unknown time [6] or in an unrecognizable
particularly recognizing abnormal user behavior in the home frequency and pattern.
[1].
There are three machine learning (ML) strategies to
Anomaly detection is a method that defines any detect anomalies, namely, supervised, unsupervised, and
irregular occurrence in a dataset [2]. In the context of a smart semi-supervised learning [7]. The selection of the technique
home, this method can be of great interest since one's daily is based on the type of dataset. Supervised learning, which
routine usually represents their physical and mental health. uses labeled data, is the methodology used in this research.
Through the monitoring of these routines, the system may ML tasks typically come in two flavors: classification and
identify deviations from standard activities, which can regression [8]. Classification is concerned with discrete
suggest changes in health or cognitive function, such as the outputs, whereas regression estimates continuous values [9].
onset of dementia [3]. This technology is most useful for Binary classification involves the model producing one of
people who do not want to stay at home but go to healthcare two values, e.g., true/false or 0/1[10]. Because this research
facilities, such as the elderly [4]. classifies behavior into either "normal" or "abnormal," it is
applicable to a binary classification problem. In this case,

IJISRT25MAR233 www.ijisrt.com 425

Volume 10, Issue 3, March – 2025 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/25mar233
"abnormal" is used to denote aberrations in smart home Ardebili et al. [16] used sensor-based data in the first
appliance and kitchen activity use. study for anomaly detection and achieved accuracy rates of
73% using SVM, 87% using RF, 79% using DT, and 68%
Various studies have investigated the use of supervised using KNN. The results showed that RF performed better
ML methods for anomaly detection in smart home settings, compared to other models in classification accuracy and thus
resulting in notable contributions to the field [11]. This study holds potential for real-world implementation. Han et al. [15]
extends prior work by tackling an important challenge: applied a vision-based sensor system for fall detection,
dependence on various activities for detecting anomalies. The comparing the same ML methods. Their findings exhibited
objective of this research is to identify unusual behavior in a greater precision, at 96% for SVM, 93% for RF, 88% for DT,
smart home from data collected from smart kitchen and 92% for KNN. The performance of vision-based systems
appliances. being better indicates their ability to detect crucial events, like
falls, in smart homes. In addition, Elhadad and Tan [17]
II. RELATED WORK employed RF and KNN-based anomaly detection system with
special emphasis on two features: time sequence and activity
Technological improvements in artificial intelligence sequence. They obtained an accuracy of 85% in their model,
and machine learning have greatly improved anomaly thus proving the relevance of sequential data in detecting
detection in smart home settings. Researchers have behavioral anomalies in smart home applications.
investigated numerous machine learning methods to examine
sensor data and detect abnormal behavior patterns. The III. METHODOLOGY
research aims to enhance the accuracy and efficiency of
abnormal activity detection, especially in health monitoring Herein, applied an anomaly detection method via
and assisted living applications. This part is a summary of the machine learning models to detect suspicious patterns in
previous research on machine learning-based anomaly home energy consumption. Our dataset has three important
detection, and its major methodologies, challenges, and features: DISHWASH, HEATHOME, and AIRCOND,
contributions. which are the most pertinent attributes to select concerning
energy usage patterns. Our method is structured around a
Jakkula and cook [12] utilized a One-Class Support pipeline format with the steps of data preprocessing,
Vector Machine (OC-SVM) for anomaly detection in home standardizing features, Isolation Forest and Local Outlier
automation, leveraging the Weka tool [13]. Sensor readings Factor (LOF)-based anomaly detection, and visualizing
were obtained from major areas, the dining room, kitchen, results. The performance of the models was tested with
and living room, to observe behavioral patterns. The OC- anomaly distribution analysis, model agreement, and
SVM model performed well, with 0.5 as the type I error rate visualization of anomaly scores.
and 1 as perfect score for precision, recall, and F-measure.
But the model was trained on normal behavior instances only,
i.e., type II error was not indicated, which may affect its
reliability in actual usage.

Likewise, Palaniappan, Bhargavi, and Vaidehi [14]

proposed a method to identify abnormal activities by
iteratively removing potential normal activities based on a
state transition table and a multi-class SVM. In Java
implementation, their SVM-based system successfully
minimized computational time using a kernel function for
training. The research incorporated nine various activities,
and in order to handle the vast amount of raw data, the sliding
window technique was utilized to divide the dataset and
regulate the data flow rate. The preprocessing procedure
helped to obtain a total accuracy of 94.4%, which testifies to
the successfulness of the presented approach in detecting
deviations in everyday activities in a smart home
environment.

Ardebili, Eken, and Küçük [15], and Han, Yang, and

Huang [16] investigated several machine learning algorithms,
such as Support Vector Machine (SVM), Random Forest
(RF), Decision Tree (DT), and k-Nearest Neighbors (KNN),
to compare their performance for anomaly detection. The
comparison among these models offered insights into system
stability and real-world applicability.

Fig 1 Poposed Methodlogy

IJISRT25MAR233 www.ijisrt.com 426

Volume 10, Issue 3, March – 2025 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/25mar233
 Dataset Description: names provide the names of the respective appliances. The
The data contains residential appliance power data comes from home load measurements on May 2, 2016.
consumption recorded at a timestamp resolution of 1Hz. The
timestamp field is the date and time in Excel's serial day  Selected Features:
format, counting days from January 1, 1900, at 12:00 AM. Three significant fetures DISHWASH, HEATHOME,
This can be reformatted to show regular date and time values and AIRCOND in this research work to study anomalies in
through Excel's number formatting. The other columns domestic energy consumption. DISHWASH indicates dish
include power consumption data for different domestic washer usage patterns, HEATHOME represents heating
appliances. Two columns are used for each appliance: one system activity, and AIRCOND represents air conditioning
column for measured values (e.g., usage. These features were used because they contribute
DP2_Condenser_Power_Val (Actual Values)) and a second significantly to overall energy consumption and have the
column that shows the unit of measurement (e.g., capability to indicate unusual patterns like irregular or over
DP2_Condenser_Power_Units). The majority of power usage, which might be indicative of malfunctioning
values are in Watts. Three special columns with VAR_Val appliances or aberrant behavior. By using anomaly detection
instead of Power_Val hold reactive power measurements, and methods, we intend to determine deviations that could be
their units are listed in VARs instead of Watts. The column used for maximizing energy efficiency and system faults
detection. Below Figure Shows the features distribution.

Fig 2 Feature Distribution of Selected Features

The below plot displays Kernel Density Estimation around 1. HEATHOME has the maximum density at the
(KDE) plots, showing the distribution of three attributes: higher peak, followed by AIRCOND and then DISHWASH.
DISHWASH, HEATHOME, and AIRCOND. Each of these The shapes of the distributions are quite comparable, pointing
attributes has a bimodal distribution, which implies two towards possible correlation or common underlying
different groups in the data. There is a smaller peak close to determinants for these features.
0 for every feature, whereas a broader, higher peak exists at

IJISRT25MAR233 www.ijisrt.com 427

Volume 10, Issue 3, March – 2025 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/25mar233

Fig 3 KDE Method

Isolation Forest (IF) is an unsupervised anomaly IV. RESULTS

detection algorithm that separates anomalies rather than
profiling typical instances. It operates by randomly choosing Diagonal plots indicate distribution of every feature
one feature and then dividing the data at a random value along where blue identifies normal data points and red does the
that feature. As anomalies are rare and dissimilar from typical anomalies. Remarkably, there are merely two anomalies
data, they are separated in fewer divisions. This renders pinpointed in each and every feature, implying the presence
Isolation Forest efficient and scalable, even for high- of sparse anomalies. The off-diagonal plots indicate scatter
dimensional datasets. The algorithm gives anomaly scores plots of each pair of features further highlighting isolation of
depending on the number of splits required to separate a point these two anomalies from other data points within the cluster.
lower score for anomalies. IF is extensively applied in fraud The Anomaly_LOF column, probably a Local Outlier Factor
detection, network security, and industrial monitoring value, clearly distinguishes between the outlier-free points
because it is fast and efficient in identifying outliers without with low values close to zero and the two outliers with higher
needing labeled data. values. Overall, the plot demonstrates how the Isolation
Forest algorithm successfully singles out these two data
points as individual outliers among the rest of the dataset,
exhibiting the algorithm's capability to identify anomalies in
a high-dimensional feature space.

Fig 4 Pairt plot for anmoly detection using isolation forest Isolation Forest detected 258 anomalies out of 5686 samples (4.54%)

IJISRT25MAR233 www.ijisrt.com 428

Volume 10, Issue 3, March – 2025 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/25mar233
The below histogram illustrates the distribution of distribution with a dense cluster of data points with scores
anomaly scores obtained from an Isolation Forest algorithm, close to 0.15, which represents normal data, and a sparse
a technique for outlier detection. The x-axis is the anomaly cluster with scores lower than the 5% level, which represents
score, with lower scores meaning higher probability of being possible anomalies. The existence of data points with scores
an anomaly. The y-axis indicates the frequency of data points well below zero indicates that the algorithm has picked up on
within each range of scores. The graph is overlaid with a obvious outliers. The 5% threshold does a good job of
Kernel Density Estimation (KDE) curve, a smoothed view of distinguishing between the bulk of the distribution and the
the distribution of scores. A dashed red vertical line indicates lower-scoring outliers, showing the algorithm's capability to
the 5% level, which is the boundary below which data points detect deviations from the typical data patterns.
are defined as anomalies. The histogram shows a bimodal

Fig 5 Vissualization of Anomly Detection Score

V. CONCLUSION ACKNOWLEDGMENT

This research proved the efficacy of the Isolation Forest Javid Gahramanov expresses gratitude to JVD Smart
(IF) algorithm in identifying anomalies in residential energy Systems for their support in facilitating this research. Naeem
consumption, particularly on three of the most significant Naseer acknowledges Muhammad Nawaz Sharif University
household appliances: dishwasher (DISHWASH), home of Agriculture, Multan, Pakistan, for providing an academic
heating (HEATHOME), and air conditioning (AIRCOND). environment conducive to this study. Ayishagul
The findings show that IF effectively detected abnormal Gahramanova extends appreciation to her colleagues in the
energy consumption patterns, which may indicate possible field of mathematics for their valuable insights. The authors
energy losses, malfunctioning appliances, or abnormal user also thank all contributors and institutions who provided data
behavior. Kernel Density Estimation (KDE) analysis and resources for this research on anomaly detection in
confirmed that the features follow a bimodal distribution, kitchen appliance usage.
which indicates different consumption behaviors. Visual
examination using pair plots and 3D visualizations also REFERENCES
confirmed that the model could identify outliers. The IF
anomaly score histogram also confirmed the results, verifying [1]. E. Hoque, R. F. Dickerson, S. M. Preum, M. A.
the robustness of the methodology. Through the use of Hanson, A. T. Barth, and J. A. Stankovic, “Holmes: A
unsupervised machine learning methods such as Isolation Comprehensive Anomaly Detection System for Daily
Forest, intelligent home monitoring systems can be used to In-home Activities,” 2015 International Conference
optimize energy efficiency, facilitate predictive maintenance, on Distributed Computing in Sensor Systems, pp. 40–
and aid in automated fault detection. Additional work may 51, 2015, [Online]. Available:
consider the integration of other appliance data, improving https://api.semanticscholar.org/CorpusID:15978673
model performance through hybrid methods, and integrating [2]. V. Chandola, A. Banerjee, and V. Kumar, “Anomaly
real-time anomaly detection to provide proactive energy detection: A survey,” ACM computing surveys
management (CSUR), vol. 41, no. 3, pp. 1–58, 2009.

IJISRT25MAR233 www.ijisrt.com 429

Volume 10, Issue 3, March – 2025 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/25mar233
[3]. K. Mengoudi et al., “Augmenting dementia cognitive Sequence Classification.,” Journal of ICT Research &
assessment with instruction-less eye-tracking tests,” Applications, vol. 17, no. 1, 2023.
IEEE J Biomed Health Inform, vol. 24, no. 11, pp.
3066–3075, 2020.
[4]. A. C. Tran, S. Marsland, J. Dietrich, H. W. Guesgen,
and P. Lyons, “Use cases for abnormal behaviour
detection in smart homes,” in Aging Friendly
Technology for Health and Independence: 8th
International Conference on Smart Homes and Health
Telematics, ICOST 2010, Seoul, Korea, June 22-24,
2010. Proceedings 8, Springer, 2010, pp. 144–151.
[5]. L. M. Dang, K. Min, H. Wang, M. J. Piran, C. H. Lee,
and H. Moon, “Sensor-based and vision-based human
activity recognition: A comprehensive survey,”
Pattern Recognit, vol. 108, p. 107561, 2020.
[6]. N. C. Tay, T. Connie, T. S. Ong, A. B. J. Teoh, and P.
S. Teh, “A review of abnormal behavior detection in
activities of daily living,” IEEE Access, vol. 11, pp.
5069–5088, 2023.
[7]. U. Fiore, F. Palmieri, A. Castiglione, and A. De Santis,
“Network anomaly detection with the restricted
Boltzmann machine,” Neurocomputing, vol. 122, pp.
13–23, 2013.
[8]. S. Omar, A. Ngadi, and H. H. Jebur, “Machine
learning techniques for anomaly detection: an
overview,” Int J Comput Appl, vol. 79, no. 2, 2013.
[9]. P. Harrington, Machine learning in action. Simon and
Schuster, 2012.
[10]. Y. Wang et al., “A survey on ambient sensor-based
abnormal behaviour detection for elderly people in
healthcare,” Electronics (Basel), vol. 12, no. 7, p.
1539, 2023.
[11]. A. B. Nassif, M. A. Talib, Q. Nasir, and F. M.
Dakalbab, “Machine learning for anomaly detection:
A systematic review,” Ieee Access, vol. 9, pp. 78658–
78700, 2021.
[12]. V. Jakkula, “Predictive data mining to learn health
vitals of a resident in a smart home,” in Seventh IEEE
International Conference on Data Mining Workshops
(ICDMW 2007), IEEE, 2007, pp. 163–168.
[13]. M. Hall, E. Frank, G. Holmes, B. Pfahringer, P.
Reutemann, and I. H. Witten, “The WEKA data
mining software: an update,” ACM SIGKDD
explorations newsletter, vol. 11, no. 1, pp. 10–18,
2009.
[14]. A. Palaniappan, R. Bhargavi, and V. Vaidehi,
“Abnormal human activity recognition using SVM
based approach,” in 2012 international conference on
recent trends in information technology, IEEE, 2012,
pp. 97–102.
[15]. K. Han, Q. Yang, and Z. Huang, “A two-stage fall
recognition algorithm based on human posture
features,” Sensors, vol. 20, no. 23, p. 6966, 2020.
[16]. E. Seyedkazemi Ardebili, S. Eken, and K. Küçük,
“Activity recognition for ambient sensing data and
rule based anomaly detection,” The International
Archives of the Photogrammetry, Remote Sensing and
Spatial Information Sciences, vol. 44, pp. 379–382,
2020.
[17]. R. M. Elhadad and Y.-F. Tan, “Anomaly Detection for
Human Home Activities Using Pattern Based

IJISRT25MAR233 www.ijisrt.com 430