IAES International Journal of Artificial Intelligence (IJ-AI)

Vol. 14, No. 2, April 2025, pp. 1252∼1259

ISSN: 2252-8938, DOI: 10.11591/ijai.v14.i2.pp1252-1259 ❒ 1252

Enhancing sepsis detection using feed-forward neural

networks with hyperparameter tuning techniques
Smitha N., Tanuja R., S. H. Manjula
Department of Computer Science and Engineering, University Visvesvaraya College of Engineering, Bangalore University,
Bengaluru, India

Article Info ABSTRACT

Article history: This paper investigates the use of feed-forward neural networks for sepsis detec-
tion, emphasizing class imbalance mitigation and hyperparameter optimization.
Received May 11, 2024
Leveraging random oversampling, synthetic minority over-sampling technique
Revised Nov 8, 2024 (SMOTE), and random sampling techniques, we address class imbalance, sig-
Accepted Nov 14, 2024 nificantly improving feed-forward neural network performance. The resulting
model achieves an impressive 83% accuracy on the test set, with notable en-
Keywords: hancements in precision, recall, and F1-score for the positive class. Hyperpa-
rameter tuning using RandomizedSearchCV identifies optimal parameters, in-
Class imbalance cluding an alpha value of 0.01 and the logistic activation function, leading to a
Feed-forward neural networks remarkable 57.5% test accuracy. GridSearchCV also contributes to model re-
GridSearchCV finement, albeit with a slightly lower test accuracy of 51.5%. These findings
Hyperparameter tuning underscore the importance of robust hyperparameter tuning methods in optimiz-
Random sampling ing feed-forward neural network models for imbalanced datasets, particularly
RandomizedSearchCV in sepsis detection. The insights gained hold promise for the development of
SMOTE more accurate diagnostic tools, ultimately improving patient outcomes in clini-
cal practice.

This is an open access article under the CC BY-SA license.

Corresponding Author:
Smitha N.
Research Scholar, Department of Computer Science and Engineering
University Visvesvaraya College of Engineering, Bangalore University
Bengaluru, India
Email: [email protected]

1. INTRODUCTION
Sepsis is a medical condition that requires immediate attention to avoid a life-threatening situation.
This is a condition arises due to body’s response to an infection. If this is not treated at the initial stage can
lead to inflammation and dysfunction of organ and finally death. Sepsis can be triggered by various infectious
agents such as bacteria, viruses, fungi, or parasites, and it can affect individuals of any age or health status.
The immune system, designed to defend against infections, can go into overdrive during sepsis, releasing an
excessive amount of inflammatory chemicals. Early identification and intervention is very much required to
avoid sepsis to mitigate to severe stages like severe sepsis and septic shock [1]. If it’s not treated in the early
stage can lead to organ failure and even death their by increasing the mortality rate. Timely intervention
includes antibiotics treatment to stabilize the initial symptoms. Sepsis remains a significant global health
challenges. Machine learning plays a crucial role in detecting sepsis by leveraging computational algorithms
and statistical models to analyze large datasets and identify patterns associated with the condition [2].
Predictive analytics can assist clinicians in identifying high-risk patients and implementing

Journal homepage: http://ijai.iaescore.com

Int J Artif Intell ISSN: 2252-8938 ❒ 1253

preventive measures. Machine learning algorithms can be integrated into clinical decision support systems
to assist healthcare professionals in diagnosing sepsis. These systems analyze real-time patient data and pro-
vide recommendations or alerts based on established sepsis detection criteria. The current state of the art in
sepsis detection and management emphasizes the importance of early identification and prompt intervention
to prevent the progression of sepsis to severe stages, including septic shock. In this work feed-forward neu-
ral network model is developed and performance is checked using different metric like accuracy, precision,
F1-score, and recall. To address class imbalance in datasets, techniques like random oversampling, synthetic
minority over-sampling technique (SMOTE), and random sampling are employed, significantly improving the
performance of feed-forward neural networks in detecting sepsis. Hyperparameter optimization methods, such
as RandomizedSearchCV and GridSearchCV, are used to fine-tune these models, leading to improved accuracy
and performance metrics. This work underscoring the importance of robust hyperparameter tuning methods in
optimizing models for imbalanced datasets for detecting sepsis.

2. LITERATURE SURVEY
According to Lydia et al. [3], after evaluating multiple machine learning models, XGBoost stood
out with a remarkable accuracy of 0.98, surpassing other models. Opting for efficiency, the study suggests
using the lazy predict library and applies random search for hyper-parameter tuning of the XGBoost classifier.
Introducing a lazy classifier to address classification challenges, the research achieves an impressive score of
0.99 using random search for optimal parameters, emphasizing the effectiveness of this approach with the
XGBoost algorithm [3].
Kijpaisalratana et al. [4] have done hyper-parameter tuning using grid search on logistic regression
model. Using a 5-fold cross-validation technique, the model’s biases are minimized. Authors have compared
the proposed model performance with the existing traditional scoring systems. They have used shapley additive
explanation (SHAP) approach to explain the role of each feature in the prediction of sepsis. But this work has a
limitation, as it’s based on international classification of diseases (ICD) code which could be wrongly classified
due to different types of error [4].
Ghias et al. [5] have used machine learning algorithms to effectively predict sepsis in the patients who
are admitted to the intensive care unit (ICU). The features used here are the vital signs of the patients who are 18
and above aged. Here, the authors have used missforest algorithm for imputation of data. No hyperparameter
tuning used, as they have used random forest, which can handle all the missing values. The XGBoost model
outperformed with 0.98 accuracy [5].
Shashikumar et al. [6] have developed an recurrent neural network (RNN) model called deep artificial
intelligence sepsis expert (DeepAISE) for early sepsis prediction. This model has achieved area under the
curve (AUC) ranging from 0.87 to 0.90. Schamoni et al. [7] have developed an ensembling neural network
(NN) to analyse medical data there by reducing error. This approach is exemplified in the early prediction of
sepsis using real intensive care unit data, showcasing its efficacy in healthcare analytics. Here, authors have
created a pool of individual learners. But they have not evaluated attribute inference attack. Through preventing
membership inference, they were able to prevent attribute inference as well [7].
A novel hybrid meta heuristic algorithm, human mental search (HMS)-particle swarm optimization
(PSO), used to optimize the deep neural network weights in early sepsis diagnosis. Here, local minima are
reduced using PSO and HMS. System outperforms with mean square error (MSE) value of 0.22 [8]. This
proposed system addresses the need to optimize the feed-forward neural network using hyper-parameter tuning,
their by showcasing the effect of imbalance dataset on the model performance and addresses the need to hyper-
parameter tuning to improve the overall performance of the model in sepsis detection [9].

3. METHOD
A feed-forward neural network is a type of artificial neural network in which information flows in one
direction: forward, from the input layer to the output layer. It consists of multiple layers of interconnected
nodes or neurons, organized into three main types of layers:
− Input layer: Input layer takes input data, and it has corresponding neurons as shown (1) and (2).

Y (1) = X (1) a + Z (1) (1)

Enhancing sepsis detection using feed-forward neural networks with hyperparameter ... (Smitha N.)
1254 ❒ ISSN: 2252-8938

a(1) = activation(Y (1) ) (2)

− Hidden layers: These are middle layers in the network. Each hidden layer has neurons and are connected
to every other neurons in the previous layer.
− Output layer: The output layer consists of number of neurons, which is completely depending on the
nature of the problem which we are solving.
A feed-forward neural network, at its core, is akin to a single-layer perceptron. It processes a sequence
of inputs by multiplying them with weights. These weighted inputs are then summed to produce a total. If this
sum surpasses a predetermined threshold (often set at zero), the output is typically 1; otherwise, it is -1. The
single-layer perceptron is widely used for classification and can also integrate machine learning features [10].

3.1. Hyperparameter tuning

Hyperparameter tuning is a pivotal aspect in refining the efficacy of feed-forward neural networks,
and GridSearchCV, RandomizedSearchCV, and random sampling techniques are commonly employed for this
purpose [11]. By oversampling the minority class randomly, it helps alleviate class imbalance issues during the
hyperparameter tuning process [12]. GridSearchCV systematically explores a predetermined hyperparameter
grid, RandomizedSearchCV samples hyperparameter combinations from specified distributions, and random
sampling involves selecting hyperparameter values at random. The incorporation of RandomOverSampler
enhances the performance, especially when class imbalances [13], [14].

3.1.1. GridSearchCV
GridSearchCV is a hyperparameter tuning technique provided by scikit-learn that exhaustively
searches through a predefined set of hyperparameter values for a given machine learning model. It performs
a cross-validated grid search over a parameter grid, evaluating the model’s performance for each combination
of hyperparameters. Hyperparameter tuning finds the best possible settings to optimize the model’s predictive
performance. In hyperparameter tuning, you define a grid of hyperparameter values that you want to explore.
This grid is essentially a set of points in the hyperparameter space [15], [6]. The grid search algorithm aims
to find the optimal hyperparameters (h∗ ) by evaluating the model’s performance across all combinations of
hyperparameter values in the grid (H) .
GridSearchCV systematically explores this grid by training and evaluating the model at each point.
It covers all possible combinations of hyperparameters specified in the grid. Cross-validation is an essential
part of the process. The model is trained on some folds and tested on others, iteratively. GridSearchCV can be
computationally expensive, especially when the hyperparameter space is large [16].

3.1.2. RandomizedSearchCV
Like GridSearchCV, the purpose of RandomizedSearchCV is hyperparameter tuning. Instead of
exhaustively searching through all possible combinations of hyperparameters, RandomizedSearchCV randomly
samples a specified number of combinations from the hyperparameter space. Similar to GridSearchCV, you
define hyperparameter space to explore. However, instead of specifying exact values, you provide distributions
from which values are randomly drawn [17].
The mathematical representation can be simplified as follows: Let H be the set of hyperparameters,
M be the machine learning model, and P be the performance metric. The randomized search algorithm aims
to find the optimal hyperparameters (h∗ ) by randomly sampling hyperparameter values from the distributions
defined in the search space, as shown in (3).

h∗ = arg max P (M (h)) (3)

h∈H

Where M (h) represents the machine learning model with hyperparameters h, P (·) is the performance metric
used for evaluation, and arg maxh∈H denotes the hyperparameter values that maximize the model’s
performance.
Hyperparameter tuning with RandomizedSearchCV is often more suitable when computational
resources are limited or when you want to quickly get a sense of the hyperparameter space [18]. In sum-
mary, RandomizedSearchCV efficiently explores the hyperparameter space through random sampling, making
it particularly useful when the search space is extensive and an exhaustive search is not feasible. It offers

Int J Artif Intell, Vol. 14, No. 2, April 2025: 1252–1259

Int J Artif Intell ISSN: 2252-8938 ❒ 1255

a good compromise between exploration and computational cost, providing an effective alternative to Grid-
SearchCV. Figure 1 shows the proposed system architecture where feed-forward neural network is optimized
using SMOTE, random sampling, GridCV, and RandomizedCV.

Figure 1. Proposed architecture

3.1.3. Random sampling

The main objective of oversampling techniques, such as RandomOverSampler, is to address class
imbalance by generating synthetic instances for the minority class. The idea is to provide the model with more
examples of the minority class, allowing it to learn a more robust decision boundary and reduce bias towards
the majority class. RandomOverSampler takes a random selection approach to create additional samples for
the minority class. Instead of relying on specific patterns within the minority class, it introduces randomness
by duplicating instances at random. The random oversampling approach helps prevent overfitting to specific
patterns in the minority class. If the oversampling were deterministic and always duplicated the same instances,
the model might memorize these patterns, leading to poor generalization to new data [19].

3.1.4. Synthetic minority over-sampling technique

SMOTE is a technique used to address class imbalance. It works by generating synthetic instances for
the minority class, and there by balancing the class. SMOTE helps balance the class there by improving the
overall performance of the model [20].

Enhancing sepsis detection using feed-forward neural networks with hyperparameter ... (Smitha N.)
1256 ❒ ISSN: 2252-8938

4. RESULTS AND DISCUSSION

In this section, it is explained the results of research and at the same time is given the comprehensive
discussion [21]. Dataset is taken from Physionet 2019 Challenge which has MIMIC III data [22]. Before
tuning, the model showed low precision (0.0869) and recall (0.5417) for the minority class, resulting in an
F1-score of 0.1497. However, after GridCV, precision significantly increased to 0.4875, achieving perfect
recall (1.0000), and an improved F1-score of 0.6555. The classification report reflected this enhancement, with
substantial gains in precision, recall, and F1-score for the minority class (1) while maintaining a reasonable
accuracy of 0.49. Subsequently, the model was also evaluated using the SMOTE technique. Before GridCV,
SMOTE demonstrated a balanced performance with precision, recall, and F1-score values around 0.57 for both
classes. After GridCV, the model’s hyperparameters were optimized, resulting in the best parameters being
’alpha’: 0.1, ’hiddenlayersizes’: (128,), and ’maxiter’: 1500.
However, despite the parameter tuning, the performance improvements were moderate, with an
accuracy of 0.51 and only slight enhancements in precision, recall, and F1-score for both classes. The SMOTE-
enhanced model reflects the challenges in further improving the classification performance beyond a certain
threshold, emphasizing the need for a comprehensive understanding of the dataset characteristics and potential
limitations in model enhancement strategies [23], [24]. The classification report obtained from the evaluation
of the feed-forward neural network with random oversampling for sepsis detection, where precision for de-
tecting sepsis (class 1) is 8.86%. In this case, out of all instances predicted as sepsis, only 8.86% were true
positive cases. Recall for detecting sepsis is 72.29%. A higher recall indicates better sensitivity. The F1-score
is 15.79%. A higher F1-score is generally desirable. The overall accuracy of the model on this dataset is 83%.
This metric represents the proportion of correctly classified instances among all instances. In the context of
sepsis detection, while the model demonstrates high accuracy, the low precision suggests a high rate of false
positives, and the F1-score indicates room for improvement in balancing precision and recall [25], [26].
Here’s an explanation for the results obtained from the RandomizedSearchCV for hyperparameter
tuning of a feed-forward neural network for sepsis detection as shown in Table 1.
− Best parameters: mlpclassifier alpha: The optimal value found by the randomized search is 0.01.
mlpclassifier activation: The logistic activation function was found to be optimal.
− Best accuracy (training set): The best accuracy achieved on the training dataset is 56.12%. This indicates
how well the neural network learned from the training data while adjusting its weights and biases based
on the provided features.
− Test accuracy of best model is obtained when evaluated on an unseen dataset (not used during
training), is 57.5%. This represents the ability of the tuned neural network to generalize and make
accurate predictions on new data, specifically for sepsis detection.

Table 1. Best parameters and accuracy for feed-forward neural network with hyperparameter tuning
Best parameters {’mlpclassifier alpha’: 0.01, ’mlpclassifier activation’: ’logistic’}
Best accuracy 0.561
Test accuracy of best model 0.575

The performance evaluation of feed-forward neural network model, employing hyperparameter

tuning techniques, namely RandomizedSearchCV and GridSearchCV, is summarized herein. The test losses
for the feed-forward neural network, obtained through RandomizedSearchCV, range from 0.0911 to 0.0931,
showcasing consistent performance across multiple iterations as shown in Figures 2 and 3. Notably, Random-
izedSearchCV yielded a test accuracy of 57.5%, with the optimal parameters being an alpha value of 0.01 and
the logistic activation function as shown in Figure 4. In contrast, the GridSearchCV approach produced test
losses within the range of 0.0915 to 0.0929, resulting in a slightly lower test accuracy of 51.5%. The optimal
parameters identified through GridSearchCV include an alpha value of 0.01 and a hidden layer size of 64.
In the context of sepsis detection, these results suggest that the feed-forward neural network,
with the specified hyperparameters, can achieve an accuracy of approximately 57.5% on unseen data. The
results obtained from the GridSearchCV for hyperparameter tuning of a feed-forward neural network for sepsis
detection as shown in Table 2 can be explained as follows: Best Parameters:alpha the optimal value found by
the grid search is 0.01. The architecture of the neural network, specifying one hidden layer with 64 neurons.
Best accuracy (training set): The best accuracy achieved on the training dataset is 52.63%. This indicates how
well the neural network learned from the training data while adjusting its weights and biases based on the pro-

Int J Artif Intell, Vol. 14, No. 2, April 2025: 1252–1259

Int J Artif Intell ISSN: 2252-8938 ❒ 1257

vided features. Test accuracy of best model is 51.5%. This represents the ability of the tuned neural network to
generalize and make accurate predictions on new data, specifically for sepsis detection. In the context of sepsis
detection, these results suggest that the feed-forward neural network, with the specified hyperparameters, can
achieve an accuracy of approximately 51.5% on unseen data.

Figure 2. Feed-forward neural network test and validation loss

Figure 3. Feed-forward neural network test loss

Figure 4. Feed-forward neural network training and test accuracy

Enhancing sepsis detection using feed-forward neural networks with hyperparameter ... (Smitha N.)
1258 ❒ ISSN: 2252-8938

Table 2. GridSearchCV results - feed-forward neural network

Best parameters {’alpha’: 0.01, ’hidden layer sizes’: (64,)}
Best accuracy 0.526
Test accuracy of best model 0.515

5. CONCLUSION
Feed-forward neural network model achieved 83% accuracy on the test set with random oversam-
pling, there is a substantial improvement in the precision, recall, and F1 score of the positive class compare to
original imbalanced dataset. Optimal hyperparameter tuning like RandomizedSearchCV identified parameters
like alpha value 0.01, logistic activation function with 57.5% accuracy. Whereas with GridSearchCV there was
a down flow accuracy which is 51.5%. These findings underscore the role of robust hyper-parameter tuning
methods in optimizing the feed-forward neural network for imbalanced datasets in detecting sepsis.

REFERENCES
[1] M. Moor, B. Rieck, M. Horn, C. R. Jutzeler, and K. Borgwardt, “Early prediction of sepsis in the icu using machine learning: a
systematic review,” Frontiers in Medicine, vol. 8, 2021, doi: 10.3389/fmed.2021.607952.
[2] S. M. Lauritsen et al., “Early detection of sepsis utilizing deep learning on electronic health record event sequences,” Artificial
Intelligence in Medicine, vol. 104, 2020, doi: 10.1016/j.artmed.2020.101820.
[3] E. L. Lydia, S. A. Althubiti, C. S. S. Anupama, and K. V. Kumar, “Prediction of sepsis disease using random search to optimize
hyperparameter tuning based on lazy predict model,” Smart Innovation, Systems and Technologies, vol. 371, pp. 351–367, 2023,
doi: 10.1007/978-981-99-6706-3 31.
[4] N. Kijpaisalratana, D. Sanglertsinlapachai, S. Techaratsami, K. Musikatavorn, and J. Saoraya, “Machine learning algorithms for
early sepsis detection in the emergency department: a retrospective study,” International Journal of Medical Informatics, vol. 160,
2022, doi: 10.1016/j.ijmedinf.2022.104689.
[5] N. Ghias et al., “Using machine learning algorithms to predict sepsis and its stages in icu patients,” medRxiv, pp. 1–9, 2022, doi:
10.1101/2022.03.15.22271655.
[6] S. P. Shashikumar, C. Josef, A. Sharma, and S. Nemati, “DeepAISE–an end-to-end development and deployment of a recurrent
neural survival model for early prediction of sepsis,” arXiv-Computer Science, pp. 1–45, 2019, doi: 10.48550/arXiv.1908.04759.
[7] S. Schamoni, M. Hagmann, and S. Riezler, “Ensembling neural networks for improved prediction and privacy in early diagnosis of
sepsis,” Proceedings of Machine Learning Research, vol. 182, pp. 123–145, 2022.
[8] U. Kaya, A. Yılmaz, and S. Aşar, “Sepsis prediction by using a hybrid metaheuristic algorithm: a novel approach for optimizing
deep neural networks,” Diagnostics, vol. 13, no. 12, 2023, doi: 10.3390/diagnostics13122023.
[9] R. R. Kanna, T. M. Priya, V. A. Immanuel, V. B. Kirubanand, T. Senthilnathan, and V. Rohini, “An novel cutting edge ann ma-
chine learning algorithm for sepsis early prediction and diagnosis,” AIP Conference Proceedings, vol. 2909, no. 1, 2023, doi:
10.1063/5.0181885.
[10] G. Bebis and M. Georgiopoulos, “Feed-forward neural networks,” IEEE Potentials, vol. 13, no. 4, pp. 27–31, 2002, doi:
10.1109/45.329294.
[11] E. Junter, “Predicting sepsis in the intensive care unit using machine learning,” M.Sc. Thesis, School of Engineering Science, LUT
University, Lappeenranta, Finlandia, 2020.
[12] Z. Wang and B. Yao, “Multi-branching temporal convolutional network for sepsis prediction,” IEEE Journal of Biomedical and
Health Informatics, vol. 26, no. 2, pp. 876–887, 2022, doi: 10.1109/JBHI.2021.3092835.
[13] R. Bardenet, M. Brendel, B. Kégl, and M. Sebag, “Collaborative hyperparameter tuning,” in 30th International Conference on
Machine Learning, ICML 2013, 2013, pp. 858–866.
[14] T. Van Steenkiste et al., “Accurate prediction of blood culture outcome in the intensive care unit using long short-term memory
neural networks,” Artificial Intelligence in Medicine, vol. 97, pp. 38–43, 2019, doi: 10.1016/j.artmed.2018.10.008.
[15] B. H. Shekar and G. Dagnew, “Grid search-based hyperparameter tuning and classification of microarray cancer data,” in 2019
Second International Conference on Advanced Computational and Communication Paradigms (ICACCP), IEEE, 2019, pp. 1–8,
doi: 10.1109/ICACCP.2019.8882943.
[16] S. Kumar, H. Tiwari, and M. Jaiswal, “Diabetes prediction using optimisation techniques with machine learning algorithms,” Inter-
national Journal of Electronic Healthcare, vol. 13, no. 2, pp. 158–168, 2023, doi: 10.1504/IJEH.2023.130515.
[17] T. N. Pham, L. Van Tran, and S. V. T. Dao, “Early disease classification of mango leaves using feed-forward neural network and
hybrid metaheuristic feature selection,” IEEE Access, vol. 8, pp. 189960–189973, 2020, doi: 10.1109/ACCESS.2020.3031914.
[18] D. L. Shanthi and N. Chethan, “Genetic algorithm based hyper-parameter tuning to improve the performance of machine learning
models,” SN Computer Science, vol. 4, no. 2, Dec. 2022, doi: 10.1007/s42979-022-01537-8.
[19] W. G. Cochran, Sampling techniques. Hoboken, USA: John Wiley & Sons, 1977, doi: 10.1002/9781118445112.stat01272.
[20] A. Fernández, S. Garcı́a, F. Herrera, and N. V. Chawla, “SMOTE for learning from imbalanced data: progress and challenges,
marking the 15-year anniversary,” Journal of Artificial Intelligence Research, vol. 61, pp. 863–905, 2018, doi: 10.1613/jair.1.11192.
[21] J. Y. Park et al., “Predicting sepsis mortality in a population based national database: machine learning approac,” Journal of Medical
Internet Research, vol. 24, no. 4, 2022.
[22] M. A. Reyna, C. S. Josef, R. Jeter, S. P. Shashikumar, M. B. Westover, S. Nemati, G. D. Clifford, and A. Sharma, “Early prediction
of sepsis from clinical data: the PhysioNet/Computing in Cardiology Challenge 2019,” Crit. Care Med., vol. 48, no. 2, pp. 210–217,
2020, doi: 10.1097/CCM.0000000000004145.
[23] P. Nejedly, F. Plesinger, I. Viscor, J. Halamek, and P. Jurak, “Prediction of sepsis using lstm with hyperparameter optimization with

Int J Artif Intell, Vol. 14, No. 2, April 2025: 1252–1259

Int J Artif Intell ISSN: 2252-8938 ❒ 1259

a genetic algorithm,” in 2019 Computing in Cardiology (CinC), Dec. 2019, pp. 1–4, doi: 10.22489/CinC.2019.022.
[24] R. Hassanzadeh, M. Farhadian, and H. Rafieemehr, “Hospital mortality prediction in traumatic injuries patients: comparing different
smote-based machine learning algorithms,” BMC Medical Research Methodology, vol. 23, no. 1, 2023, doi: 10.1186/s12874-023-
01920-w.
[25] I. Persson, A. Östling, M. Arlbrandt, J. Söderberg, and D. Becedas, “A machine learning sepsis prediction algorithm for intended
intensive care unit use (navoy sepsis): proof-of-concept study,” JMIR Formative Research, vol. 5, no. 9, 2021, doi: 10.2196/28000.
[26] E. K. Sari and D. P. Sari, “The effect of mobile learning application in enhancing students’ knowledge of caring sepsis patients,”
International Journal of Public Health Science, vol. 12, no. 2, pp. 787–793, 2023, doi: 10.11591/ijphs.v12i2.22206.

BIOGRAPHIES OF AUTHORS

Smitha N. is Research Scholar (Bangalore University), Department of Computer Science

and Engineering, University Visvesvaraya College of Engineering, Bengaluru. Her research interests
are in the field of machine learning, deep learning, and natural language processing. She can be
contacted at email: [email protected]

Tanuja R. is Associate Professor in the Department of Computer Science and Engineering

at University Visvesvaraya College of Engineering. She was awarded Ph.D. in Computer Science and
Engineering from Bangalore University. Her research interests are in the field of computer networks,
network security, and machine learning. She has around 21 years of teaching experience, published
32 papers in international conferences and refereed journals. She guiding 6 Ph.D. research scholars.
She can be contacted at email: [email protected].

S. H. Manjula currently a Professor in the Department of Computer Science & Engineering

at the University Visvesvaraya College of Engineering, Bangalore University, Bengaluru. She has
established herself as a prominent figure in the field. Holding a B.E., M.Tech., and Ph.D. in Computer
Science & Engineering, she has honed her expertise in computer networks, wireless sensor networks,
data mining, cloud computing, artificial intelligence, machine learning, and federated learning. With
an impressive track record, she has authored 69 journals, presented 72 conference papers, holds 5
patents, and has contributed to 4 book chapters while publishing 4 books, showcasing her prolific and
diverse contributions to the academic and technological community. Additionally, she is currently a
respected member of the Executive Council and has served as a former member of the Academic
Senate at VTU Belagavi. She can be contacted at email: [email protected].

Enhancing sepsis detection using feed-forward neural networks with hyperparameter ... (Smitha N.)