Enhanced Skill Optimization Algorithm and Stacked Long Short-Term Memory with Sech Activation Function for Gastrointestinal Disease

Janagama Srividya; Harikrishna Bommala

doi:10.2478/ijssis-2026-0030

.blurhash-client-img { display: none !important; }

Enhanced Skill Optimization Algorithm and Stacked Long Short-Term Memory with Sech Activation Function for Gastrointestinal Disease

International Journal on Smart Sensing and Intelligent Systems

Volume 19 (2026): Issue 1 (January 2026)

By: Janagama Srividya and Harikrishna Bommala

Open Access

|May 2026

Figures & Tables

Workflow of proposed methodology. AHE, adaptive histogram equalization; ESOA, enhanced skill optimization algorithm; Stacked LSTM-SAF, stacked long short-term memory with Sech activation function.

Workflow of the ESOA method. ESOA, enhanced skill optimization algorithm.

Structure of the stacked LSTM. Stacked LSTM, stacked long short-term memory.

Sample images of actual and predicted labels (A) Kvasir-V1 (B) Kvasir-V2 dataset.

Evaluation of confusion matrix for Stacked LSTM-SAF (A) Kvasir-V1, (B) Kvasir-V2, (C) HyperKvasir. Stacked LSTM-SAF, stacked long short-term memory with Sech activation function.

Evaluation of RoC curve for Stacked LSTM-SAF: (A) Kvasir-V1, (B) Kvasir-V2, (C) HyperKvasir. ROC, receiver operating characteristics; Stacked LSTM-SAF, stacked long short-term memory with Sech activation function.

Analysis of standard deviation for the proposed method: (A) Kvasir-V1, (B) Kvasir-V2, (C) HyperKvasir.

Performance analysis of different feature selection methods

Methods	Accuracy (%)	Precision (%)	Recall (%)	F1-score (%)	Specificity (%)
Kvasir-V1
OOA	94.17	94.54	93.92	94.23	92.67
COA	96.66	95.15	95.38	95.27	94.38
BES	97.37	95.12	96.30	95.71	94.79
SOA	97.99	96.53	97.98	97.25	96.95
ESOA	99.60	99.20	98.71	98.96	99.88

Kvasir-V2

OOA	94.26	94.65	93.33	94.65	93.33
COA	95.58	95.21	94.67	95.21	94.67
BES	97.51	95.18	96.84	95.18	96.84
SOA	97.51	97.18	96.84	97.18	96.84
ESOA	99.88	99.61	97.12	99.61	97.12

HyperKvasir

OOA	94.32	94.78	93.85	94.11	92.91
COA	96.28	95.42	95.16	95.29	94.63
BES	97.14	95.88	96.41	96.14	95.37
SOA	97.92	96.71	97.65	97.18	96.82
ESOA	99.74	99.33	98.95	99.14	99.61

Performance analysis of different classification methods

Methods	Accuracy (%)	Precision (%)	Recall (%)	F1-score (%)
Kvasir-V1
RNN	89.44	90.51	88.90	91.42
GRU	93.80	94.03	93.59	92.14
LSTM	96.04	95.99	94.88	94.66
Stacked-LSTM	99.60	98.71	99.88	99.20

Kvasir-V2

RNN	90.56	90.50	91.27	92.95
GRU	94.07	93.10	92.00	94.56
LSTM	97.49	96.40	95.55	95.61
Stacked-LSTM	99.88	97.93	97.12	99.61

HyperKvasir

RNN	89.88	90.65	89.44	91.12
GRU	93.92	94.21	93.78	92.83
LSTM	96.38	96.10	95.21	95.68
Stacked-LSTM	99.74	98.95	99.33	99.14

Performance analysis of computational complexity across datasets

Methods	Datasets	Memory consumption (MB)	Training time (s)	Inference time (s)
RNN	Kvasir-V1	27.12	37.95	36.74
GRU		23.05	32.58	30.84
LSTM		8.67	30.58	26.59
Stacked-LSTM-SAF		6.72	2.54	9.58

RNN	Kvasir-V2	27.48	38.62	37.12
GRU		22.87	35.12	31.04
LSTM		9.02	31.28	27.11
Stacked-LSTM-SAF		6.58	2.73	9.92

RNN	HyperKvasir	30.48	42.69	40.98
GRU		27.46	25.96	22.02
LSTM		21.69	19.78	17.36
Stacked-LSTM-SAF		8.46	5.80	11.96

Comparative analysis of existing methods on Kvasir-V1 and V2 datasets

Methods	Datasets	Accuracy (%)	Precision (%)	Recall (%)	F1-score (%)
LPNet [17]	Kvasir-V1	93.55	93.55	93.55	93.55
VGG16 kernle RBF [19]	Kvasir-V2	96.64	97	97	97
SK-Net [20]	Kvasir-V1	98.45	N/A	96.60	N/A
	Kvasir-V2	97.83	N/A	N/A	N/A
Star-GAN + InceptionNet-V3 [21]	Kvasir-V2	94.96	N/A	94.93	94.93
CapsNet [22]	Kvasir-V2	93.40	N/A	N/A	N/A
Proposed ESOA with Stacked LSTM-SAF	Kvasir-V1	99.60	98.71	99.88	99.20
Proposed ESOA with Stacked LSTM-SAF	Kvasir-V2	99.88	97.93	97.12	99.61

Cross-dataset validation results: Trained on Kvasir-V1 and tested on Kvasir-V2

Methods	Accuracy (%)	Precision (%)	Recall (%)	F1-score (%)
RNN	87.35	88.20	86.50	87.34
GRU	91.12	92.05	90.50	91.27
LSTM	94.50	95.10	93.80	94.44
Stacked LSTM-SAF	97.25	96.80	96.00	96.39

Performance analysis of different activation functions

Methods	Accuracy (%)	Precision (%)	Recall (%)	F1-score (%)	T-test from p-values	CI (%)
Kvasir-V1
Stacked LSTM-ReLU	94.97	93.47	92.60	93.63	0.032	86.18
Stacked LSTM-Tanh	95.46	93.28	94.30	95.04	0.030	87.63
Stacked LSTM-Sigmoid	97.03	96.96	95.43	96.01	0.029	89.06
Stacked LSTM-sech	99.50	98.71	99.88	99.20	0.026	94.12

Kvasir-V2

Stacked LSTM-ReLU	93.55	92.53	92.67	91.76	0.034	89.60
Stacked LSTM-Tanh	96.62	95.71	96.00	97.26	0.030	90.17
Stacked LSTM-Sigmoid	97.45	95.50	94.18	96.65	0.028	91.78
Stacked LSTM-sech	99.88	97.93	97.12	99.61	0.024	94.36

HyperKvasir

Stacked LSTM-ReLU	94.76	93.62	92.89	93.45	0.036	87.15
Stacked LSTM-Tanh	95.84	94.71	95.25	95.07	0.034	89.36
Stacked LSTM-Sigmoid	97.28	96.64	95.81	96.22	0.031	91.58
Stacked LSTM-Sech	99.74	98.95	99.33	99.14	0.029	94.85

References

Authors

Metrics

Articles in this issue

DOI: https://doi.org/10.2478/ijssis-2026-0030 | Journal eISSN: 1178-5608

Journal RSS Feed

Language: English

Submitted on: Aug 7, 2025

Published on: May 28, 2026

Published by: Macquarie University, Australia

In partnership with: Paradigm Publishing Services

Publication frequency: 1 issue per year

Keywords:

deep learning,

enhanced skill optimization algorithm,

gastrointestinal disease,

Gorilla troop optimizer,

stacked long short-term memory with Sech activation function

Related subjects:

Engineering,

Introductions and overviews,

Engineering, other

© 2026 Janagama Srividya, Harikrishna Bommala, published by Macquarie University, Australia
This work is licensed under the Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License.

Volume 19 (2026): Issue 1 (January 2026)

Enhanced Skill Optimization Algorithm and Stacked Long Short-Term Memory with Sech Activation Function for Gastrointestinal Disease

Figures & Tables

Figure 1:

Figure 2:

Figure 3:

Figure 4:

Figure 5:

Figure 6:

Figure 7:

Figure 8:

Performance analysis of different feature selection methods

Performance analysis of different classification methods

Performance analysis of computational complexity across datasets

Comparative analysis of existing methods on Kvasir-V1 and V2 datasets

Cross-dataset validation results: Trained on Kvasir-V1 and tested on Kvasir-V2

Performance analysis of different activation functions

Paradigm

My account